4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie.
7 * kswapd added: 7.1.96 sct
8 * Removed kswapd_ctl limits, and swap out as many pages as needed
9 * to bring the system back to freepages.high: 2.4.97, Rik van Riel.
10 * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
11 * Multiqueue VM started 5.8.00, Rik van Riel.
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/module.h>
18 #include <linux/gfp.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/swap.h>
21 #include <linux/pagemap.h>
22 #include <linux/init.h>
23 #include <linux/highmem.h>
24 #include <linux/vmpressure.h>
25 #include <linux/vmstat.h>
26 #include <linux/file.h>
27 #include <linux/writeback.h>
28 #include <linux/blkdev.h>
29 #include <linux/buffer_head.h> /* for try_to_release_page(),
30 buffer_heads_over_limit */
31 #include <linux/mm_inline.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rmap.h>
34 #include <linux/topology.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/compaction.h>
38 #include <linux/notifier.h>
39 #include <linux/rwsem.h>
40 #include <linux/delay.h>
41 #include <linux/kthread.h>
42 #include <linux/freezer.h>
43 #include <linux/memcontrol.h>
44 #include <linux/delayacct.h>
45 #include <linux/sysctl.h>
46 #include <linux/oom.h>
47 #include <linux/prefetch.h>
48 #include <linux/printk.h>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
53 #include <linux/swapops.h>
54 #include <linux/balloon_compaction.h>
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/vmscan.h>
62 /* How many pages shrink_list() should reclaim */
63 unsigned long nr_to_reclaim
;
65 /* This context's GFP mask */
68 /* Allocation order */
72 * Nodemask of nodes allowed by the caller. If NULL, all nodes
78 * The memory cgroup that hit its limit and as a result is the
79 * primary target of this reclaim invocation.
81 struct mem_cgroup
*target_mem_cgroup
;
83 /* Scan (total_size >> priority) pages at once */
86 unsigned int may_writepage
:1;
88 /* Can mapped pages be reclaimed? */
89 unsigned int may_unmap
:1;
91 /* Can pages be swapped as part of reclaim? */
92 unsigned int may_swap
:1;
94 unsigned int hibernation_mode
:1;
96 /* One of the zones is ready for compaction */
97 unsigned int compaction_ready
:1;
99 /* Incremented by the number of inactive pages that were scanned */
100 unsigned long nr_scanned
;
102 /* Number of pages freed so far during a call to shrink_zones() */
103 unsigned long nr_reclaimed
;
106 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
108 #ifdef ARCH_HAS_PREFETCH
109 #define prefetch_prev_lru_page(_page, _base, _field) \
111 if ((_page)->lru.prev != _base) { \
114 prev = lru_to_page(&(_page->lru)); \
115 prefetch(&prev->_field); \
119 #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
122 #ifdef ARCH_HAS_PREFETCHW
123 #define prefetchw_prev_lru_page(_page, _base, _field) \
125 if ((_page)->lru.prev != _base) { \
128 prev = lru_to_page(&(_page->lru)); \
129 prefetchw(&prev->_field); \
133 #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
137 * From 0 .. 100. Higher means more swappy.
139 int vm_swappiness
= 60;
141 * The total number of pages which are beyond the high watermark within all
144 unsigned long vm_total_pages
;
146 static LIST_HEAD(shrinker_list
);
147 static DECLARE_RWSEM(shrinker_rwsem
);
150 static bool global_reclaim(struct scan_control
*sc
)
152 return !sc
->target_mem_cgroup
;
155 static bool global_reclaim(struct scan_control
*sc
)
161 static unsigned long zone_reclaimable_pages(struct zone
*zone
)
165 nr
= zone_page_state(zone
, NR_ACTIVE_FILE
) +
166 zone_page_state(zone
, NR_INACTIVE_FILE
);
168 if (get_nr_swap_pages() > 0)
169 nr
+= zone_page_state(zone
, NR_ACTIVE_ANON
) +
170 zone_page_state(zone
, NR_INACTIVE_ANON
);
175 bool zone_reclaimable(struct zone
*zone
)
177 return zone_page_state(zone
, NR_PAGES_SCANNED
) <
178 zone_reclaimable_pages(zone
) * 6;
181 static unsigned long get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
183 if (!mem_cgroup_disabled())
184 return mem_cgroup_get_lru_size(lruvec
, lru
);
186 return zone_page_state(lruvec_zone(lruvec
), NR_LRU_BASE
+ lru
);
190 * Add a shrinker callback to be called from the vm.
192 int register_shrinker(struct shrinker
*shrinker
)
194 size_t size
= sizeof(*shrinker
->nr_deferred
);
197 * If we only have one possible node in the system anyway, save
198 * ourselves the trouble and disable NUMA aware behavior. This way we
199 * will save memory and some small loop time later.
201 if (nr_node_ids
== 1)
202 shrinker
->flags
&= ~SHRINKER_NUMA_AWARE
;
204 if (shrinker
->flags
& SHRINKER_NUMA_AWARE
)
207 shrinker
->nr_deferred
= kzalloc(size
, GFP_KERNEL
);
208 if (!shrinker
->nr_deferred
)
211 down_write(&shrinker_rwsem
);
212 list_add_tail(&shrinker
->list
, &shrinker_list
);
213 up_write(&shrinker_rwsem
);
216 EXPORT_SYMBOL(register_shrinker
);
221 void unregister_shrinker(struct shrinker
*shrinker
)
223 down_write(&shrinker_rwsem
);
224 list_del(&shrinker
->list
);
225 up_write(&shrinker_rwsem
);
226 kfree(shrinker
->nr_deferred
);
228 EXPORT_SYMBOL(unregister_shrinker
);
230 #define SHRINK_BATCH 128
233 shrink_slab_node(struct shrink_control
*shrinkctl
, struct shrinker
*shrinker
,
234 unsigned long nr_pages_scanned
, unsigned long lru_pages
)
236 unsigned long freed
= 0;
237 unsigned long long delta
;
242 int nid
= shrinkctl
->nid
;
243 long batch_size
= shrinker
->batch
? shrinker
->batch
246 freeable
= shrinker
->count_objects(shrinker
, shrinkctl
);
251 * copy the current shrinker scan count into a local variable
252 * and zero it so that other concurrent shrinker invocations
253 * don't also do this scanning work.
255 nr
= atomic_long_xchg(&shrinker
->nr_deferred
[nid
], 0);
258 delta
= (4 * nr_pages_scanned
) / shrinker
->seeks
;
260 do_div(delta
, lru_pages
+ 1);
262 if (total_scan
< 0) {
263 pr_err("shrink_slab: %pF negative objects to delete nr=%ld\n",
264 shrinker
->scan_objects
, total_scan
);
265 total_scan
= freeable
;
269 * We need to avoid excessive windup on filesystem shrinkers
270 * due to large numbers of GFP_NOFS allocations causing the
271 * shrinkers to return -1 all the time. This results in a large
272 * nr being built up so when a shrink that can do some work
273 * comes along it empties the entire cache due to nr >>>
274 * freeable. This is bad for sustaining a working set in
277 * Hence only allow the shrinker to scan the entire cache when
278 * a large delta change is calculated directly.
280 if (delta
< freeable
/ 4)
281 total_scan
= min(total_scan
, freeable
/ 2);
284 * Avoid risking looping forever due to too large nr value:
285 * never try to free more than twice the estimate number of
288 if (total_scan
> freeable
* 2)
289 total_scan
= freeable
* 2;
291 trace_mm_shrink_slab_start(shrinker
, shrinkctl
, nr
,
292 nr_pages_scanned
, lru_pages
,
293 freeable
, delta
, total_scan
);
296 * Normally, we should not scan less than batch_size objects in one
297 * pass to avoid too frequent shrinker calls, but if the slab has less
298 * than batch_size objects in total and we are really tight on memory,
299 * we will try to reclaim all available objects, otherwise we can end
300 * up failing allocations although there are plenty of reclaimable
301 * objects spread over several slabs with usage less than the
304 * We detect the "tight on memory" situations by looking at the total
305 * number of objects we want to scan (total_scan). If it is greater
306 * than the total number of objects on slab (freeable), we must be
307 * scanning at high prio and therefore should try to reclaim as much as
310 while (total_scan
>= batch_size
||
311 total_scan
>= freeable
) {
313 unsigned long nr_to_scan
= min(batch_size
, total_scan
);
315 shrinkctl
->nr_to_scan
= nr_to_scan
;
316 ret
= shrinker
->scan_objects(shrinker
, shrinkctl
);
317 if (ret
== SHRINK_STOP
)
321 count_vm_events(SLABS_SCANNED
, nr_to_scan
);
322 total_scan
-= nr_to_scan
;
328 * move the unused scan count back into the shrinker in a
329 * manner that handles concurrent updates. If we exhausted the
330 * scan, there is no need to do an update.
333 new_nr
= atomic_long_add_return(total_scan
,
334 &shrinker
->nr_deferred
[nid
]);
336 new_nr
= atomic_long_read(&shrinker
->nr_deferred
[nid
]);
338 trace_mm_shrink_slab_end(shrinker
, nid
, freed
, nr
, new_nr
, total_scan
);
343 * Call the shrink functions to age shrinkable caches
345 * Here we assume it costs one seek to replace a lru page and that it also
346 * takes a seek to recreate a cache object. With this in mind we age equal
347 * percentages of the lru and ageable caches. This should balance the seeks
348 * generated by these structures.
350 * If the vm encountered mapped pages on the LRU it increase the pressure on
351 * slab to avoid swapping.
353 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
355 * `lru_pages' represents the number of on-LRU pages in all the zones which
356 * are eligible for the caller's allocation attempt. It is used for balancing
357 * slab reclaim versus page reclaim.
359 * Returns the number of slab objects which we shrunk.
361 unsigned long shrink_slab(struct shrink_control
*shrinkctl
,
362 unsigned long nr_pages_scanned
,
363 unsigned long lru_pages
)
365 struct shrinker
*shrinker
;
366 unsigned long freed
= 0;
368 if (nr_pages_scanned
== 0)
369 nr_pages_scanned
= SWAP_CLUSTER_MAX
;
371 if (!down_read_trylock(&shrinker_rwsem
)) {
373 * If we would return 0, our callers would understand that we
374 * have nothing else to shrink and give up trying. By returning
375 * 1 we keep it going and assume we'll be able to shrink next
382 list_for_each_entry(shrinker
, &shrinker_list
, list
) {
383 if (!(shrinker
->flags
& SHRINKER_NUMA_AWARE
)) {
385 freed
+= shrink_slab_node(shrinkctl
, shrinker
,
386 nr_pages_scanned
, lru_pages
);
390 for_each_node_mask(shrinkctl
->nid
, shrinkctl
->nodes_to_scan
) {
391 if (node_online(shrinkctl
->nid
))
392 freed
+= shrink_slab_node(shrinkctl
, shrinker
,
393 nr_pages_scanned
, lru_pages
);
397 up_read(&shrinker_rwsem
);
403 static inline int is_page_cache_freeable(struct page
*page
)
406 * A freeable page cache page is referenced only by the caller
407 * that isolated the page, the page cache radix tree and
408 * optional buffer heads at page->private.
410 return page_count(page
) - page_has_private(page
) == 2;
413 static int may_write_to_queue(struct backing_dev_info
*bdi
,
414 struct scan_control
*sc
)
416 if (current
->flags
& PF_SWAPWRITE
)
418 if (!bdi_write_congested(bdi
))
420 if (bdi
== current
->backing_dev_info
)
426 * We detected a synchronous write error writing a page out. Probably
427 * -ENOSPC. We need to propagate that into the address_space for a subsequent
428 * fsync(), msync() or close().
430 * The tricky part is that after writepage we cannot touch the mapping: nothing
431 * prevents it from being freed up. But we have a ref on the page and once
432 * that page is locked, the mapping is pinned.
434 * We're allowed to run sleeping lock_page() here because we know the caller has
437 static void handle_write_error(struct address_space
*mapping
,
438 struct page
*page
, int error
)
441 if (page_mapping(page
) == mapping
)
442 mapping_set_error(mapping
, error
);
446 /* possible outcome of pageout() */
448 /* failed to write page out, page is locked */
450 /* move page to the active list, page is locked */
452 /* page has been sent to the disk successfully, page is unlocked */
454 /* page is clean and locked */
459 * pageout is called by shrink_page_list() for each dirty page.
460 * Calls ->writepage().
462 static pageout_t
pageout(struct page
*page
, struct address_space
*mapping
,
463 struct scan_control
*sc
)
466 * If the page is dirty, only perform writeback if that write
467 * will be non-blocking. To prevent this allocation from being
468 * stalled by pagecache activity. But note that there may be
469 * stalls if we need to run get_block(). We could test
470 * PagePrivate for that.
472 * If this process is currently in __generic_file_write_iter() against
473 * this page's queue, we can perform writeback even if that
476 * If the page is swapcache, write it back even if that would
477 * block, for some throttling. This happens by accident, because
478 * swap_backing_dev_info is bust: it doesn't reflect the
479 * congestion state of the swapdevs. Easy to fix, if needed.
481 if (!is_page_cache_freeable(page
))
485 * Some data journaling orphaned pages can have
486 * page->mapping == NULL while being dirty with clean buffers.
488 if (page_has_private(page
)) {
489 if (try_to_free_buffers(page
)) {
490 ClearPageDirty(page
);
491 pr_info("%s: orphaned page\n", __func__
);
497 if (mapping
->a_ops
->writepage
== NULL
)
498 return PAGE_ACTIVATE
;
499 if (!may_write_to_queue(mapping
->backing_dev_info
, sc
))
502 if (clear_page_dirty_for_io(page
)) {
504 struct writeback_control wbc
= {
505 .sync_mode
= WB_SYNC_NONE
,
506 .nr_to_write
= SWAP_CLUSTER_MAX
,
508 .range_end
= LLONG_MAX
,
512 SetPageReclaim(page
);
513 res
= mapping
->a_ops
->writepage(page
, &wbc
);
515 handle_write_error(mapping
, page
, res
);
516 if (res
== AOP_WRITEPAGE_ACTIVATE
) {
517 ClearPageReclaim(page
);
518 return PAGE_ACTIVATE
;
521 if (!PageWriteback(page
)) {
522 /* synchronous write or broken a_ops? */
523 ClearPageReclaim(page
);
525 trace_mm_vmscan_writepage(page
, trace_reclaim_flags(page
));
526 inc_zone_page_state(page
, NR_VMSCAN_WRITE
);
534 * Same as remove_mapping, but if the page is removed from the mapping, it
535 * gets returned with a refcount of 0.
537 static int __remove_mapping(struct address_space
*mapping
, struct page
*page
,
540 BUG_ON(!PageLocked(page
));
541 BUG_ON(mapping
!= page_mapping(page
));
543 spin_lock_irq(&mapping
->tree_lock
);
545 * The non racy check for a busy page.
547 * Must be careful with the order of the tests. When someone has
548 * a ref to the page, it may be possible that they dirty it then
549 * drop the reference. So if PageDirty is tested before page_count
550 * here, then the following race may occur:
552 * get_user_pages(&page);
553 * [user mapping goes away]
555 * !PageDirty(page) [good]
556 * SetPageDirty(page);
558 * !page_count(page) [good, discard it]
560 * [oops, our write_to data is lost]
562 * Reversing the order of the tests ensures such a situation cannot
563 * escape unnoticed. The smp_rmb is needed to ensure the page->flags
564 * load is not satisfied before that of page->_count.
566 * Note that if SetPageDirty is always performed via set_page_dirty,
567 * and thus under tree_lock, then this ordering is not required.
569 if (!page_freeze_refs(page
, 2))
571 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
572 if (unlikely(PageDirty(page
))) {
573 page_unfreeze_refs(page
, 2);
577 if (PageSwapCache(page
)) {
578 swp_entry_t swap
= { .val
= page_private(page
) };
579 mem_cgroup_swapout(page
, swap
);
580 __delete_from_swap_cache(page
);
581 spin_unlock_irq(&mapping
->tree_lock
);
582 swapcache_free(swap
);
584 void (*freepage
)(struct page
*);
587 freepage
= mapping
->a_ops
->freepage
;
589 * Remember a shadow entry for reclaimed file cache in
590 * order to detect refaults, thus thrashing, later on.
592 * But don't store shadows in an address space that is
593 * already exiting. This is not just an optizimation,
594 * inode reclaim needs to empty out the radix tree or
595 * the nodes are lost. Don't plant shadows behind its
598 if (reclaimed
&& page_is_file_cache(page
) &&
599 !mapping_exiting(mapping
))
600 shadow
= workingset_eviction(mapping
, page
);
601 __delete_from_page_cache(page
, shadow
);
602 spin_unlock_irq(&mapping
->tree_lock
);
604 if (freepage
!= NULL
)
611 spin_unlock_irq(&mapping
->tree_lock
);
616 * Attempt to detach a locked page from its ->mapping. If it is dirty or if
617 * someone else has a ref on the page, abort and return 0. If it was
618 * successfully detached, return 1. Assumes the caller has a single ref on
621 int remove_mapping(struct address_space
*mapping
, struct page
*page
)
623 if (__remove_mapping(mapping
, page
, false)) {
625 * Unfreezing the refcount with 1 rather than 2 effectively
626 * drops the pagecache ref for us without requiring another
629 page_unfreeze_refs(page
, 1);
636 * putback_lru_page - put previously isolated page onto appropriate LRU list
637 * @page: page to be put back to appropriate lru list
639 * Add previously isolated @page to appropriate LRU list.
640 * Page may still be unevictable for other reasons.
642 * lru_lock must not be held, interrupts must be enabled.
644 void putback_lru_page(struct page
*page
)
647 int was_unevictable
= PageUnevictable(page
);
649 VM_BUG_ON_PAGE(PageLRU(page
), page
);
652 ClearPageUnevictable(page
);
654 if (page_evictable(page
)) {
656 * For evictable pages, we can use the cache.
657 * In event of a race, worst case is we end up with an
658 * unevictable page on [in]active list.
659 * We know how to handle that.
661 is_unevictable
= false;
665 * Put unevictable pages directly on zone's unevictable
668 is_unevictable
= true;
669 add_page_to_unevictable_list(page
);
671 * When racing with an mlock or AS_UNEVICTABLE clearing
672 * (page is unlocked) make sure that if the other thread
673 * does not observe our setting of PG_lru and fails
674 * isolation/check_move_unevictable_pages,
675 * we see PG_mlocked/AS_UNEVICTABLE cleared below and move
676 * the page back to the evictable list.
678 * The other side is TestClearPageMlocked() or shmem_lock().
684 * page's status can change while we move it among lru. If an evictable
685 * page is on unevictable list, it never be freed. To avoid that,
686 * check after we added it to the list, again.
688 if (is_unevictable
&& page_evictable(page
)) {
689 if (!isolate_lru_page(page
)) {
693 /* This means someone else dropped this page from LRU
694 * So, it will be freed or putback to LRU again. There is
695 * nothing to do here.
699 if (was_unevictable
&& !is_unevictable
)
700 count_vm_event(UNEVICTABLE_PGRESCUED
);
701 else if (!was_unevictable
&& is_unevictable
)
702 count_vm_event(UNEVICTABLE_PGCULLED
);
704 put_page(page
); /* drop ref from isolate */
707 enum page_references
{
709 PAGEREF_RECLAIM_CLEAN
,
714 static enum page_references
page_check_references(struct page
*page
,
715 struct scan_control
*sc
)
717 int referenced_ptes
, referenced_page
;
718 unsigned long vm_flags
;
720 referenced_ptes
= page_referenced(page
, 1, sc
->target_mem_cgroup
,
722 referenced_page
= TestClearPageReferenced(page
);
725 * Mlock lost the isolation race with us. Let try_to_unmap()
726 * move the page to the unevictable list.
728 if (vm_flags
& VM_LOCKED
)
729 return PAGEREF_RECLAIM
;
731 if (referenced_ptes
) {
732 if (PageSwapBacked(page
))
733 return PAGEREF_ACTIVATE
;
735 * All mapped pages start out with page table
736 * references from the instantiating fault, so we need
737 * to look twice if a mapped file page is used more
740 * Mark it and spare it for another trip around the
741 * inactive list. Another page table reference will
742 * lead to its activation.
744 * Note: the mark is set for activated pages as well
745 * so that recently deactivated but used pages are
748 SetPageReferenced(page
);
750 if (referenced_page
|| referenced_ptes
> 1)
751 return PAGEREF_ACTIVATE
;
754 * Activate file-backed executable pages after first usage.
756 if (vm_flags
& VM_EXEC
)
757 return PAGEREF_ACTIVATE
;
762 /* Reclaim if clean, defer dirty pages to writeback */
763 if (referenced_page
&& !PageSwapBacked(page
))
764 return PAGEREF_RECLAIM_CLEAN
;
766 return PAGEREF_RECLAIM
;
769 /* Check if a page is dirty or under writeback */
770 static void page_check_dirty_writeback(struct page
*page
,
771 bool *dirty
, bool *writeback
)
773 struct address_space
*mapping
;
776 * Anonymous pages are not handled by flushers and must be written
777 * from reclaim context. Do not stall reclaim based on them
779 if (!page_is_file_cache(page
)) {
785 /* By default assume that the page flags are accurate */
786 *dirty
= PageDirty(page
);
787 *writeback
= PageWriteback(page
);
789 /* Verify dirty/writeback state if the filesystem supports it */
790 if (!page_has_private(page
))
793 mapping
= page_mapping(page
);
794 if (mapping
&& mapping
->a_ops
->is_dirty_writeback
)
795 mapping
->a_ops
->is_dirty_writeback(page
, dirty
, writeback
);
799 * shrink_page_list() returns the number of reclaimed pages
801 static unsigned long shrink_page_list(struct list_head
*page_list
,
803 struct scan_control
*sc
,
804 enum ttu_flags ttu_flags
,
805 unsigned long *ret_nr_dirty
,
806 unsigned long *ret_nr_unqueued_dirty
,
807 unsigned long *ret_nr_congested
,
808 unsigned long *ret_nr_writeback
,
809 unsigned long *ret_nr_immediate
,
812 LIST_HEAD(ret_pages
);
813 LIST_HEAD(free_pages
);
815 unsigned long nr_unqueued_dirty
= 0;
816 unsigned long nr_dirty
= 0;
817 unsigned long nr_congested
= 0;
818 unsigned long nr_reclaimed
= 0;
819 unsigned long nr_writeback
= 0;
820 unsigned long nr_immediate
= 0;
824 while (!list_empty(page_list
)) {
825 struct address_space
*mapping
;
828 enum page_references references
= PAGEREF_RECLAIM_CLEAN
;
829 bool dirty
, writeback
;
833 page
= lru_to_page(page_list
);
834 list_del(&page
->lru
);
836 if (!trylock_page(page
))
839 VM_BUG_ON_PAGE(PageActive(page
), page
);
840 VM_BUG_ON_PAGE(page_zone(page
) != zone
, page
);
844 if (unlikely(!page_evictable(page
)))
847 if (!sc
->may_unmap
&& page_mapped(page
))
850 /* Double the slab pressure for mapped and swapcache pages */
851 if (page_mapped(page
) || PageSwapCache(page
))
854 may_enter_fs
= (sc
->gfp_mask
& __GFP_FS
) ||
855 (PageSwapCache(page
) && (sc
->gfp_mask
& __GFP_IO
));
858 * The number of dirty pages determines if a zone is marked
859 * reclaim_congested which affects wait_iff_congested. kswapd
860 * will stall and start writing pages if the tail of the LRU
861 * is all dirty unqueued pages.
863 page_check_dirty_writeback(page
, &dirty
, &writeback
);
864 if (dirty
|| writeback
)
867 if (dirty
&& !writeback
)
871 * Treat this page as congested if the underlying BDI is or if
872 * pages are cycling through the LRU so quickly that the
873 * pages marked for immediate reclaim are making it to the
874 * end of the LRU a second time.
876 mapping
= page_mapping(page
);
877 if ((mapping
&& bdi_write_congested(mapping
->backing_dev_info
)) ||
878 (writeback
&& PageReclaim(page
)))
882 * If a page at the tail of the LRU is under writeback, there
883 * are three cases to consider.
885 * 1) If reclaim is encountering an excessive number of pages
886 * under writeback and this page is both under writeback and
887 * PageReclaim then it indicates that pages are being queued
888 * for IO but are being recycled through the LRU before the
889 * IO can complete. Waiting on the page itself risks an
890 * indefinite stall if it is impossible to writeback the
891 * page due to IO error or disconnected storage so instead
892 * note that the LRU is being scanned too quickly and the
893 * caller can stall after page list has been processed.
895 * 2) Global reclaim encounters a page, memcg encounters a
896 * page that is not marked for immediate reclaim or
897 * the caller does not have __GFP_IO. In this case mark
898 * the page for immediate reclaim and continue scanning.
900 * __GFP_IO is checked because a loop driver thread might
901 * enter reclaim, and deadlock if it waits on a page for
902 * which it is needed to do the write (loop masks off
903 * __GFP_IO|__GFP_FS for this reason); but more thought
904 * would probably show more reasons.
906 * Don't require __GFP_FS, since we're not going into the
907 * FS, just waiting on its writeback completion. Worryingly,
908 * ext4 gfs2 and xfs allocate pages with
909 * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing
910 * may_enter_fs here is liable to OOM on them.
912 * 3) memcg encounters a page that is not already marked
913 * PageReclaim. memcg does not have any dirty pages
914 * throttling so we could easily OOM just because too many
915 * pages are in writeback and there is nothing else to
916 * reclaim. Wait for the writeback to complete.
918 if (PageWriteback(page
)) {
920 if (current_is_kswapd() &&
922 test_bit(ZONE_WRITEBACK
, &zone
->flags
)) {
927 } else if (global_reclaim(sc
) ||
928 !PageReclaim(page
) || !(sc
->gfp_mask
& __GFP_IO
)) {
930 * This is slightly racy - end_page_writeback()
931 * might have just cleared PageReclaim, then
932 * setting PageReclaim here end up interpreted
933 * as PageReadahead - but that does not matter
934 * enough to care. What we do want is for this
935 * page to have PageReclaim set next time memcg
936 * reclaim reaches the tests above, so it will
937 * then wait_on_page_writeback() to avoid OOM;
938 * and it's also appropriate in global reclaim.
940 SetPageReclaim(page
);
947 wait_on_page_writeback(page
);
952 references
= page_check_references(page
, sc
);
954 switch (references
) {
955 case PAGEREF_ACTIVATE
:
956 goto activate_locked
;
959 case PAGEREF_RECLAIM
:
960 case PAGEREF_RECLAIM_CLEAN
:
961 ; /* try to reclaim the page below */
965 * Anonymous process memory has backing store?
966 * Try to allocate it some swap space here.
968 if (PageAnon(page
) && !PageSwapCache(page
)) {
969 if (!(sc
->gfp_mask
& __GFP_IO
))
971 if (!add_to_swap(page
, page_list
))
972 goto activate_locked
;
975 /* Adding to swap updated mapping */
976 mapping
= page_mapping(page
);
980 * The page is mapped into the page tables of one or more
981 * processes. Try to unmap it here.
983 if (page_mapped(page
) && mapping
) {
984 switch (try_to_unmap(page
, ttu_flags
)) {
986 goto activate_locked
;
992 ; /* try to free the page below */
996 if (PageDirty(page
)) {
998 * Only kswapd can writeback filesystem pages to
999 * avoid risk of stack overflow but only writeback
1000 * if many dirty pages have been encountered.
1002 if (page_is_file_cache(page
) &&
1003 (!current_is_kswapd() ||
1004 !test_bit(ZONE_DIRTY
, &zone
->flags
))) {
1006 * Immediately reclaim when written back.
1007 * Similar in principal to deactivate_page()
1008 * except we already have the page isolated
1009 * and know it's dirty
1011 inc_zone_page_state(page
, NR_VMSCAN_IMMEDIATE
);
1012 SetPageReclaim(page
);
1017 if (references
== PAGEREF_RECLAIM_CLEAN
)
1021 if (!sc
->may_writepage
)
1024 /* Page is dirty, try to write it out here */
1025 switch (pageout(page
, mapping
, sc
)) {
1029 goto activate_locked
;
1031 if (PageWriteback(page
))
1033 if (PageDirty(page
))
1037 * A synchronous write - probably a ramdisk. Go
1038 * ahead and try to reclaim the page.
1040 if (!trylock_page(page
))
1042 if (PageDirty(page
) || PageWriteback(page
))
1044 mapping
= page_mapping(page
);
1046 ; /* try to free the page below */
1051 * If the page has buffers, try to free the buffer mappings
1052 * associated with this page. If we succeed we try to free
1055 * We do this even if the page is PageDirty().
1056 * try_to_release_page() does not perform I/O, but it is
1057 * possible for a page to have PageDirty set, but it is actually
1058 * clean (all its buffers are clean). This happens if the
1059 * buffers were written out directly, with submit_bh(). ext3
1060 * will do this, as well as the blockdev mapping.
1061 * try_to_release_page() will discover that cleanness and will
1062 * drop the buffers and mark the page clean - it can be freed.
1064 * Rarely, pages can have buffers and no ->mapping. These are
1065 * the pages which were not successfully invalidated in
1066 * truncate_complete_page(). We try to drop those buffers here
1067 * and if that worked, and the page is no longer mapped into
1068 * process address space (page_count == 1) it can be freed.
1069 * Otherwise, leave the page on the LRU so it is swappable.
1071 if (page_has_private(page
)) {
1072 if (!try_to_release_page(page
, sc
->gfp_mask
))
1073 goto activate_locked
;
1074 if (!mapping
&& page_count(page
) == 1) {
1076 if (put_page_testzero(page
))
1080 * rare race with speculative reference.
1081 * the speculative reference will free
1082 * this page shortly, so we may
1083 * increment nr_reclaimed here (and
1084 * leave it off the LRU).
1092 if (!mapping
|| !__remove_mapping(mapping
, page
, true))
1096 * At this point, we have no other references and there is
1097 * no way to pick any more up (removed from LRU, removed
1098 * from pagecache). Can use non-atomic bitops now (and
1099 * we obviously don't have to worry about waking up a process
1100 * waiting on the page lock, because there are no references.
1102 __clear_page_locked(page
);
1107 * Is there need to periodically free_page_list? It would
1108 * appear not as the counts should be low
1110 list_add(&page
->lru
, &free_pages
);
1114 if (PageSwapCache(page
))
1115 try_to_free_swap(page
);
1117 putback_lru_page(page
);
1121 /* Not a candidate for swapping, so reclaim swap space. */
1122 if (PageSwapCache(page
) && vm_swap_full())
1123 try_to_free_swap(page
);
1124 VM_BUG_ON_PAGE(PageActive(page
), page
);
1125 SetPageActive(page
);
1130 list_add(&page
->lru
, &ret_pages
);
1131 VM_BUG_ON_PAGE(PageLRU(page
) || PageUnevictable(page
), page
);
1134 mem_cgroup_uncharge_list(&free_pages
);
1135 free_hot_cold_page_list(&free_pages
, true);
1137 list_splice(&ret_pages
, page_list
);
1138 count_vm_events(PGACTIVATE
, pgactivate
);
1140 *ret_nr_dirty
+= nr_dirty
;
1141 *ret_nr_congested
+= nr_congested
;
1142 *ret_nr_unqueued_dirty
+= nr_unqueued_dirty
;
1143 *ret_nr_writeback
+= nr_writeback
;
1144 *ret_nr_immediate
+= nr_immediate
;
1145 return nr_reclaimed
;
1148 unsigned long reclaim_clean_pages_from_list(struct zone
*zone
,
1149 struct list_head
*page_list
)
1151 struct scan_control sc
= {
1152 .gfp_mask
= GFP_KERNEL
,
1153 .priority
= DEF_PRIORITY
,
1156 unsigned long ret
, dummy1
, dummy2
, dummy3
, dummy4
, dummy5
;
1157 struct page
*page
, *next
;
1158 LIST_HEAD(clean_pages
);
1160 list_for_each_entry_safe(page
, next
, page_list
, lru
) {
1161 if (page_is_file_cache(page
) && !PageDirty(page
) &&
1162 !isolated_balloon_page(page
)) {
1163 ClearPageActive(page
);
1164 list_move(&page
->lru
, &clean_pages
);
1168 ret
= shrink_page_list(&clean_pages
, zone
, &sc
,
1169 TTU_UNMAP
|TTU_IGNORE_ACCESS
,
1170 &dummy1
, &dummy2
, &dummy3
, &dummy4
, &dummy5
, true);
1171 list_splice(&clean_pages
, page_list
);
1172 mod_zone_page_state(zone
, NR_ISOLATED_FILE
, -ret
);
1177 * Attempt to remove the specified page from its LRU. Only take this page
1178 * if it is of the appropriate PageActive status. Pages which are being
1179 * freed elsewhere are also ignored.
1181 * page: page to consider
1182 * mode: one of the LRU isolation modes defined above
1184 * returns 0 on success, -ve errno on failure.
1186 int __isolate_lru_page(struct page
*page
, isolate_mode_t mode
)
1190 /* Only take pages on the LRU. */
1194 /* Compaction should not handle unevictable pages but CMA can do so */
1195 if (PageUnevictable(page
) && !(mode
& ISOLATE_UNEVICTABLE
))
1201 * To minimise LRU disruption, the caller can indicate that it only
1202 * wants to isolate pages it will be able to operate on without
1203 * blocking - clean pages for the most part.
1205 * ISOLATE_CLEAN means that only clean pages should be isolated. This
1206 * is used by reclaim when it is cannot write to backing storage
1208 * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages
1209 * that it is possible to migrate without blocking
1211 if (mode
& (ISOLATE_CLEAN
|ISOLATE_ASYNC_MIGRATE
)) {
1212 /* All the caller can do on PageWriteback is block */
1213 if (PageWriteback(page
))
1216 if (PageDirty(page
)) {
1217 struct address_space
*mapping
;
1219 /* ISOLATE_CLEAN means only clean pages */
1220 if (mode
& ISOLATE_CLEAN
)
1224 * Only pages without mappings or that have a
1225 * ->migratepage callback are possible to migrate
1228 mapping
= page_mapping(page
);
1229 if (mapping
&& !mapping
->a_ops
->migratepage
)
1234 if ((mode
& ISOLATE_UNMAPPED
) && page_mapped(page
))
1237 if (likely(get_page_unless_zero(page
))) {
1239 * Be careful not to clear PageLRU until after we're
1240 * sure the page is not being freed elsewhere -- the
1241 * page release code relies on it.
1251 * zone->lru_lock is heavily contended. Some of the functions that
1252 * shrink the lists perform better by taking out a batch of pages
1253 * and working on them outside the LRU lock.
1255 * For pagecache intensive workloads, this function is the hottest
1256 * spot in the kernel (apart from copy_*_user functions).
1258 * Appropriate locks must be held before calling this function.
1260 * @nr_to_scan: The number of pages to look through on the list.
1261 * @lruvec: The LRU vector to pull pages from.
1262 * @dst: The temp list to put pages on to.
1263 * @nr_scanned: The number of pages that were scanned.
1264 * @sc: The scan_control struct for this reclaim session
1265 * @mode: One of the LRU isolation modes
1266 * @lru: LRU list id for isolating
1268 * returns how many pages were moved onto *@dst.
1270 static unsigned long isolate_lru_pages(unsigned long nr_to_scan
,
1271 struct lruvec
*lruvec
, struct list_head
*dst
,
1272 unsigned long *nr_scanned
, struct scan_control
*sc
,
1273 isolate_mode_t mode
, enum lru_list lru
)
1275 struct list_head
*src
= &lruvec
->lists
[lru
];
1276 unsigned long nr_taken
= 0;
1279 for (scan
= 0; scan
< nr_to_scan
&& !list_empty(src
); scan
++) {
1283 page
= lru_to_page(src
);
1284 prefetchw_prev_lru_page(page
, src
, flags
);
1286 VM_BUG_ON_PAGE(!PageLRU(page
), page
);
1288 switch (__isolate_lru_page(page
, mode
)) {
1290 nr_pages
= hpage_nr_pages(page
);
1291 mem_cgroup_update_lru_size(lruvec
, lru
, -nr_pages
);
1292 list_move(&page
->lru
, dst
);
1293 nr_taken
+= nr_pages
;
1297 /* else it is being freed elsewhere */
1298 list_move(&page
->lru
, src
);
1307 trace_mm_vmscan_lru_isolate(sc
->order
, nr_to_scan
, scan
,
1308 nr_taken
, mode
, is_file_lru(lru
));
1313 * isolate_lru_page - tries to isolate a page from its LRU list
1314 * @page: page to isolate from its LRU list
1316 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1317 * vmstat statistic corresponding to whatever LRU list the page was on.
1319 * Returns 0 if the page was removed from an LRU list.
1320 * Returns -EBUSY if the page was not on an LRU list.
1322 * The returned page will have PageLRU() cleared. If it was found on
1323 * the active list, it will have PageActive set. If it was found on
1324 * the unevictable list, it will have the PageUnevictable bit set. That flag
1325 * may need to be cleared by the caller before letting the page go.
1327 * The vmstat statistic corresponding to the list on which the page was
1328 * found will be decremented.
1331 * (1) Must be called with an elevated refcount on the page. This is a
1332 * fundamentnal difference from isolate_lru_pages (which is called
1333 * without a stable reference).
1334 * (2) the lru_lock must not be held.
1335 * (3) interrupts must be enabled.
1337 int isolate_lru_page(struct page
*page
)
1341 VM_BUG_ON_PAGE(!page_count(page
), page
);
1343 if (PageLRU(page
)) {
1344 struct zone
*zone
= page_zone(page
);
1345 struct lruvec
*lruvec
;
1347 spin_lock_irq(&zone
->lru_lock
);
1348 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
1349 if (PageLRU(page
)) {
1350 int lru
= page_lru(page
);
1353 del_page_from_lru_list(page
, lruvec
, lru
);
1356 spin_unlock_irq(&zone
->lru_lock
);
1362 * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and
1363 * then get resheduled. When there are massive number of tasks doing page
1364 * allocation, such sleeping direct reclaimers may keep piling up on each CPU,
1365 * the LRU list will go small and be scanned faster than necessary, leading to
1366 * unnecessary swapping, thrashing and OOM.
1368 static int too_many_isolated(struct zone
*zone
, int file
,
1369 struct scan_control
*sc
)
1371 unsigned long inactive
, isolated
;
1373 if (current_is_kswapd())
1376 if (!global_reclaim(sc
))
1380 inactive
= zone_page_state(zone
, NR_INACTIVE_FILE
);
1381 isolated
= zone_page_state(zone
, NR_ISOLATED_FILE
);
1383 inactive
= zone_page_state(zone
, NR_INACTIVE_ANON
);
1384 isolated
= zone_page_state(zone
, NR_ISOLATED_ANON
);
1388 * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they
1389 * won't get blocked by normal direct-reclaimers, forming a circular
1392 if ((sc
->gfp_mask
& GFP_IOFS
) == GFP_IOFS
)
1395 return isolated
> inactive
;
1398 static noinline_for_stack
void
1399 putback_inactive_pages(struct lruvec
*lruvec
, struct list_head
*page_list
)
1401 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
1402 struct zone
*zone
= lruvec_zone(lruvec
);
1403 LIST_HEAD(pages_to_free
);
1406 * Put back any unfreeable pages.
1408 while (!list_empty(page_list
)) {
1409 struct page
*page
= lru_to_page(page_list
);
1412 VM_BUG_ON_PAGE(PageLRU(page
), page
);
1413 list_del(&page
->lru
);
1414 if (unlikely(!page_evictable(page
))) {
1415 spin_unlock_irq(&zone
->lru_lock
);
1416 putback_lru_page(page
);
1417 spin_lock_irq(&zone
->lru_lock
);
1421 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
1424 lru
= page_lru(page
);
1425 add_page_to_lru_list(page
, lruvec
, lru
);
1427 if (is_active_lru(lru
)) {
1428 int file
= is_file_lru(lru
);
1429 int numpages
= hpage_nr_pages(page
);
1430 reclaim_stat
->recent_rotated
[file
] += numpages
;
1432 if (put_page_testzero(page
)) {
1433 __ClearPageLRU(page
);
1434 __ClearPageActive(page
);
1435 del_page_from_lru_list(page
, lruvec
, lru
);
1437 if (unlikely(PageCompound(page
))) {
1438 spin_unlock_irq(&zone
->lru_lock
);
1439 mem_cgroup_uncharge(page
);
1440 (*get_compound_page_dtor(page
))(page
);
1441 spin_lock_irq(&zone
->lru_lock
);
1443 list_add(&page
->lru
, &pages_to_free
);
1448 * To save our caller's stack, now use input list for pages to free.
1450 list_splice(&pages_to_free
, page_list
);
1454 * If a kernel thread (such as nfsd for loop-back mounts) services
1455 * a backing device by writing to the page cache it sets PF_LESS_THROTTLE.
1456 * In that case we should only throttle if the backing device it is
1457 * writing to is congested. In other cases it is safe to throttle.
1459 static int current_may_throttle(void)
1461 return !(current
->flags
& PF_LESS_THROTTLE
) ||
1462 current
->backing_dev_info
== NULL
||
1463 bdi_write_congested(current
->backing_dev_info
);
1467 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1468 * of reclaimed pages
1470 static noinline_for_stack
unsigned long
1471 shrink_inactive_list(unsigned long nr_to_scan
, struct lruvec
*lruvec
,
1472 struct scan_control
*sc
, enum lru_list lru
)
1474 LIST_HEAD(page_list
);
1475 unsigned long nr_scanned
;
1476 unsigned long nr_reclaimed
= 0;
1477 unsigned long nr_taken
;
1478 unsigned long nr_dirty
= 0;
1479 unsigned long nr_congested
= 0;
1480 unsigned long nr_unqueued_dirty
= 0;
1481 unsigned long nr_writeback
= 0;
1482 unsigned long nr_immediate
= 0;
1483 isolate_mode_t isolate_mode
= 0;
1484 int file
= is_file_lru(lru
);
1485 struct zone
*zone
= lruvec_zone(lruvec
);
1486 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
1488 while (unlikely(too_many_isolated(zone
, file
, sc
))) {
1489 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
1491 /* We are about to die and free our memory. Return now. */
1492 if (fatal_signal_pending(current
))
1493 return SWAP_CLUSTER_MAX
;
1499 isolate_mode
|= ISOLATE_UNMAPPED
;
1500 if (!sc
->may_writepage
)
1501 isolate_mode
|= ISOLATE_CLEAN
;
1503 spin_lock_irq(&zone
->lru_lock
);
1505 nr_taken
= isolate_lru_pages(nr_to_scan
, lruvec
, &page_list
,
1506 &nr_scanned
, sc
, isolate_mode
, lru
);
1508 __mod_zone_page_state(zone
, NR_LRU_BASE
+ lru
, -nr_taken
);
1509 __mod_zone_page_state(zone
, NR_ISOLATED_ANON
+ file
, nr_taken
);
1511 if (global_reclaim(sc
)) {
1512 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, nr_scanned
);
1513 if (current_is_kswapd())
1514 __count_zone_vm_events(PGSCAN_KSWAPD
, zone
, nr_scanned
);
1516 __count_zone_vm_events(PGSCAN_DIRECT
, zone
, nr_scanned
);
1518 spin_unlock_irq(&zone
->lru_lock
);
1523 nr_reclaimed
= shrink_page_list(&page_list
, zone
, sc
, TTU_UNMAP
,
1524 &nr_dirty
, &nr_unqueued_dirty
, &nr_congested
,
1525 &nr_writeback
, &nr_immediate
,
1528 spin_lock_irq(&zone
->lru_lock
);
1530 reclaim_stat
->recent_scanned
[file
] += nr_taken
;
1532 if (global_reclaim(sc
)) {
1533 if (current_is_kswapd())
1534 __count_zone_vm_events(PGSTEAL_KSWAPD
, zone
,
1537 __count_zone_vm_events(PGSTEAL_DIRECT
, zone
,
1541 putback_inactive_pages(lruvec
, &page_list
);
1543 __mod_zone_page_state(zone
, NR_ISOLATED_ANON
+ file
, -nr_taken
);
1545 spin_unlock_irq(&zone
->lru_lock
);
1547 mem_cgroup_uncharge_list(&page_list
);
1548 free_hot_cold_page_list(&page_list
, true);
1551 * If reclaim is isolating dirty pages under writeback, it implies
1552 * that the long-lived page allocation rate is exceeding the page
1553 * laundering rate. Either the global limits are not being effective
1554 * at throttling processes due to the page distribution throughout
1555 * zones or there is heavy usage of a slow backing device. The
1556 * only option is to throttle from reclaim context which is not ideal
1557 * as there is no guarantee the dirtying process is throttled in the
1558 * same way balance_dirty_pages() manages.
1560 * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number
1561 * of pages under pages flagged for immediate reclaim and stall if any
1562 * are encountered in the nr_immediate check below.
1564 if (nr_writeback
&& nr_writeback
== nr_taken
)
1565 set_bit(ZONE_WRITEBACK
, &zone
->flags
);
1568 * memcg will stall in page writeback so only consider forcibly
1569 * stalling for global reclaim
1571 if (global_reclaim(sc
)) {
1573 * Tag a zone as congested if all the dirty pages scanned were
1574 * backed by a congested BDI and wait_iff_congested will stall.
1576 if (nr_dirty
&& nr_dirty
== nr_congested
)
1577 set_bit(ZONE_CONGESTED
, &zone
->flags
);
1580 * If dirty pages are scanned that are not queued for IO, it
1581 * implies that flushers are not keeping up. In this case, flag
1582 * the zone ZONE_DIRTY and kswapd will start writing pages from
1585 if (nr_unqueued_dirty
== nr_taken
)
1586 set_bit(ZONE_DIRTY
, &zone
->flags
);
1589 * If kswapd scans pages marked marked for immediate
1590 * reclaim and under writeback (nr_immediate), it implies
1591 * that pages are cycling through the LRU faster than
1592 * they are written so also forcibly stall.
1594 if (nr_immediate
&& current_may_throttle())
1595 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
1599 * Stall direct reclaim for IO completions if underlying BDIs or zone
1600 * is congested. Allow kswapd to continue until it starts encountering
1601 * unqueued dirty pages or cycling through the LRU too quickly.
1603 if (!sc
->hibernation_mode
&& !current_is_kswapd() &&
1604 current_may_throttle())
1605 wait_iff_congested(zone
, BLK_RW_ASYNC
, HZ
/10);
1607 trace_mm_vmscan_lru_shrink_inactive(zone
->zone_pgdat
->node_id
,
1609 nr_scanned
, nr_reclaimed
,
1611 trace_shrink_flags(file
));
1612 return nr_reclaimed
;
1616 * This moves pages from the active list to the inactive list.
1618 * We move them the other way if the page is referenced by one or more
1619 * processes, from rmap.
1621 * If the pages are mostly unmapped, the processing is fast and it is
1622 * appropriate to hold zone->lru_lock across the whole operation. But if
1623 * the pages are mapped, the processing is slow (page_referenced()) so we
1624 * should drop zone->lru_lock around each page. It's impossible to balance
1625 * this, so instead we remove the pages from the LRU while processing them.
1626 * It is safe to rely on PG_active against the non-LRU pages in here because
1627 * nobody will play with that bit on a non-LRU page.
1629 * The downside is that we have to touch page->_count against each page.
1630 * But we had to alter page->flags anyway.
1633 static void move_active_pages_to_lru(struct lruvec
*lruvec
,
1634 struct list_head
*list
,
1635 struct list_head
*pages_to_free
,
1638 struct zone
*zone
= lruvec_zone(lruvec
);
1639 unsigned long pgmoved
= 0;
1643 while (!list_empty(list
)) {
1644 page
= lru_to_page(list
);
1645 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
1647 VM_BUG_ON_PAGE(PageLRU(page
), page
);
1650 nr_pages
= hpage_nr_pages(page
);
1651 mem_cgroup_update_lru_size(lruvec
, lru
, nr_pages
);
1652 list_move(&page
->lru
, &lruvec
->lists
[lru
]);
1653 pgmoved
+= nr_pages
;
1655 if (put_page_testzero(page
)) {
1656 __ClearPageLRU(page
);
1657 __ClearPageActive(page
);
1658 del_page_from_lru_list(page
, lruvec
, lru
);
1660 if (unlikely(PageCompound(page
))) {
1661 spin_unlock_irq(&zone
->lru_lock
);
1662 mem_cgroup_uncharge(page
);
1663 (*get_compound_page_dtor(page
))(page
);
1664 spin_lock_irq(&zone
->lru_lock
);
1666 list_add(&page
->lru
, pages_to_free
);
1669 __mod_zone_page_state(zone
, NR_LRU_BASE
+ lru
, pgmoved
);
1670 if (!is_active_lru(lru
))
1671 __count_vm_events(PGDEACTIVATE
, pgmoved
);
1674 static void shrink_active_list(unsigned long nr_to_scan
,
1675 struct lruvec
*lruvec
,
1676 struct scan_control
*sc
,
1679 unsigned long nr_taken
;
1680 unsigned long nr_scanned
;
1681 unsigned long vm_flags
;
1682 LIST_HEAD(l_hold
); /* The pages which were snipped off */
1683 LIST_HEAD(l_active
);
1684 LIST_HEAD(l_inactive
);
1686 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
1687 unsigned long nr_rotated
= 0;
1688 isolate_mode_t isolate_mode
= 0;
1689 int file
= is_file_lru(lru
);
1690 struct zone
*zone
= lruvec_zone(lruvec
);
1695 isolate_mode
|= ISOLATE_UNMAPPED
;
1696 if (!sc
->may_writepage
)
1697 isolate_mode
|= ISOLATE_CLEAN
;
1699 spin_lock_irq(&zone
->lru_lock
);
1701 nr_taken
= isolate_lru_pages(nr_to_scan
, lruvec
, &l_hold
,
1702 &nr_scanned
, sc
, isolate_mode
, lru
);
1703 if (global_reclaim(sc
))
1704 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, nr_scanned
);
1706 reclaim_stat
->recent_scanned
[file
] += nr_taken
;
1708 __count_zone_vm_events(PGREFILL
, zone
, nr_scanned
);
1709 __mod_zone_page_state(zone
, NR_LRU_BASE
+ lru
, -nr_taken
);
1710 __mod_zone_page_state(zone
, NR_ISOLATED_ANON
+ file
, nr_taken
);
1711 spin_unlock_irq(&zone
->lru_lock
);
1713 while (!list_empty(&l_hold
)) {
1715 page
= lru_to_page(&l_hold
);
1716 list_del(&page
->lru
);
1718 if (unlikely(!page_evictable(page
))) {
1719 putback_lru_page(page
);
1723 if (unlikely(buffer_heads_over_limit
)) {
1724 if (page_has_private(page
) && trylock_page(page
)) {
1725 if (page_has_private(page
))
1726 try_to_release_page(page
, 0);
1731 if (page_referenced(page
, 0, sc
->target_mem_cgroup
,
1733 nr_rotated
+= hpage_nr_pages(page
);
1735 * Identify referenced, file-backed active pages and
1736 * give them one more trip around the active list. So
1737 * that executable code get better chances to stay in
1738 * memory under moderate memory pressure. Anon pages
1739 * are not likely to be evicted by use-once streaming
1740 * IO, plus JVM can create lots of anon VM_EXEC pages,
1741 * so we ignore them here.
1743 if ((vm_flags
& VM_EXEC
) && page_is_file_cache(page
)) {
1744 list_add(&page
->lru
, &l_active
);
1749 ClearPageActive(page
); /* we are de-activating */
1750 list_add(&page
->lru
, &l_inactive
);
1754 * Move pages back to the lru list.
1756 spin_lock_irq(&zone
->lru_lock
);
1758 * Count referenced pages from currently used mappings as rotated,
1759 * even though only some of them are actually re-activated. This
1760 * helps balance scan pressure between file and anonymous pages in
1763 reclaim_stat
->recent_rotated
[file
] += nr_rotated
;
1765 move_active_pages_to_lru(lruvec
, &l_active
, &l_hold
, lru
);
1766 move_active_pages_to_lru(lruvec
, &l_inactive
, &l_hold
, lru
- LRU_ACTIVE
);
1767 __mod_zone_page_state(zone
, NR_ISOLATED_ANON
+ file
, -nr_taken
);
1768 spin_unlock_irq(&zone
->lru_lock
);
1770 mem_cgroup_uncharge_list(&l_hold
);
1771 free_hot_cold_page_list(&l_hold
, true);
1775 static int inactive_anon_is_low_global(struct zone
*zone
)
1777 unsigned long active
, inactive
;
1779 active
= zone_page_state(zone
, NR_ACTIVE_ANON
);
1780 inactive
= zone_page_state(zone
, NR_INACTIVE_ANON
);
1782 if (inactive
* zone
->inactive_ratio
< active
)
1789 * inactive_anon_is_low - check if anonymous pages need to be deactivated
1790 * @lruvec: LRU vector to check
1792 * Returns true if the zone does not have enough inactive anon pages,
1793 * meaning some active anon pages need to be deactivated.
1795 static int inactive_anon_is_low(struct lruvec
*lruvec
)
1798 * If we don't have swap space, anonymous page deactivation
1801 if (!total_swap_pages
)
1804 if (!mem_cgroup_disabled())
1805 return mem_cgroup_inactive_anon_is_low(lruvec
);
1807 return inactive_anon_is_low_global(lruvec_zone(lruvec
));
1810 static inline int inactive_anon_is_low(struct lruvec
*lruvec
)
1817 * inactive_file_is_low - check if file pages need to be deactivated
1818 * @lruvec: LRU vector to check
1820 * When the system is doing streaming IO, memory pressure here
1821 * ensures that active file pages get deactivated, until more
1822 * than half of the file pages are on the inactive list.
1824 * Once we get to that situation, protect the system's working
1825 * set from being evicted by disabling active file page aging.
1827 * This uses a different ratio than the anonymous pages, because
1828 * the page cache uses a use-once replacement algorithm.
1830 static int inactive_file_is_low(struct lruvec
*lruvec
)
1832 unsigned long inactive
;
1833 unsigned long active
;
1835 inactive
= get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1836 active
= get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1838 return active
> inactive
;
1841 static int inactive_list_is_low(struct lruvec
*lruvec
, enum lru_list lru
)
1843 if (is_file_lru(lru
))
1844 return inactive_file_is_low(lruvec
);
1846 return inactive_anon_is_low(lruvec
);
1849 static unsigned long shrink_list(enum lru_list lru
, unsigned long nr_to_scan
,
1850 struct lruvec
*lruvec
, struct scan_control
*sc
)
1852 if (is_active_lru(lru
)) {
1853 if (inactive_list_is_low(lruvec
, lru
))
1854 shrink_active_list(nr_to_scan
, lruvec
, sc
, lru
);
1858 return shrink_inactive_list(nr_to_scan
, lruvec
, sc
, lru
);
1869 * Determine how aggressively the anon and file LRU lists should be
1870 * scanned. The relative value of each set of LRU lists is determined
1871 * by looking at the fraction of the pages scanned we did rotate back
1872 * onto the active list instead of evict.
1874 * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan
1875 * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan
1877 static void get_scan_count(struct lruvec
*lruvec
, int swappiness
,
1878 struct scan_control
*sc
, unsigned long *nr
)
1880 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
1882 u64 denominator
= 0; /* gcc */
1883 struct zone
*zone
= lruvec_zone(lruvec
);
1884 unsigned long anon_prio
, file_prio
;
1885 enum scan_balance scan_balance
;
1886 unsigned long anon
, file
;
1887 bool force_scan
= false;
1888 unsigned long ap
, fp
;
1894 * If the zone or memcg is small, nr[l] can be 0. This
1895 * results in no scanning on this priority and a potential
1896 * priority drop. Global direct reclaim can go to the next
1897 * zone and tends to have no problems. Global kswapd is for
1898 * zone balancing and it needs to scan a minimum amount. When
1899 * reclaiming for a memcg, a priority drop can cause high
1900 * latencies, so it's better to scan a minimum amount there as
1903 if (current_is_kswapd() && !zone_reclaimable(zone
))
1905 if (!global_reclaim(sc
))
1908 /* If we have no swap space, do not bother scanning anon pages. */
1909 if (!sc
->may_swap
|| (get_nr_swap_pages() <= 0)) {
1910 scan_balance
= SCAN_FILE
;
1915 * Global reclaim will swap to prevent OOM even with no
1916 * swappiness, but memcg users want to use this knob to
1917 * disable swapping for individual groups completely when
1918 * using the memory controller's swap limit feature would be
1921 if (!global_reclaim(sc
) && !swappiness
) {
1922 scan_balance
= SCAN_FILE
;
1927 * Do not apply any pressure balancing cleverness when the
1928 * system is close to OOM, scan both anon and file equally
1929 * (unless the swappiness setting disagrees with swapping).
1931 if (!sc
->priority
&& swappiness
) {
1932 scan_balance
= SCAN_EQUAL
;
1937 * Prevent the reclaimer from falling into the cache trap: as
1938 * cache pages start out inactive, every cache fault will tip
1939 * the scan balance towards the file LRU. And as the file LRU
1940 * shrinks, so does the window for rotation from references.
1941 * This means we have a runaway feedback loop where a tiny
1942 * thrashing file LRU becomes infinitely more attractive than
1943 * anon pages. Try to detect this based on file LRU size.
1945 if (global_reclaim(sc
)) {
1946 unsigned long zonefile
;
1947 unsigned long zonefree
;
1949 zonefree
= zone_page_state(zone
, NR_FREE_PAGES
);
1950 zonefile
= zone_page_state(zone
, NR_ACTIVE_FILE
) +
1951 zone_page_state(zone
, NR_INACTIVE_FILE
);
1953 if (unlikely(zonefile
+ zonefree
<= high_wmark_pages(zone
))) {
1954 scan_balance
= SCAN_ANON
;
1960 * There is enough inactive page cache, do not reclaim
1961 * anything from the anonymous working set right now.
1963 if (!inactive_file_is_low(lruvec
)) {
1964 scan_balance
= SCAN_FILE
;
1968 scan_balance
= SCAN_FRACT
;
1971 * With swappiness at 100, anonymous and file have the same priority.
1972 * This scanning priority is essentially the inverse of IO cost.
1974 anon_prio
= swappiness
;
1975 file_prio
= 200 - anon_prio
;
1978 * OK, so we have swap space and a fair amount of page cache
1979 * pages. We use the recently rotated / recently scanned
1980 * ratios to determine how valuable each cache is.
1982 * Because workloads change over time (and to avoid overflow)
1983 * we keep these statistics as a floating average, which ends
1984 * up weighing recent references more than old ones.
1986 * anon in [0], file in [1]
1989 anon
= get_lru_size(lruvec
, LRU_ACTIVE_ANON
) +
1990 get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1991 file
= get_lru_size(lruvec
, LRU_ACTIVE_FILE
) +
1992 get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1994 spin_lock_irq(&zone
->lru_lock
);
1995 if (unlikely(reclaim_stat
->recent_scanned
[0] > anon
/ 4)) {
1996 reclaim_stat
->recent_scanned
[0] /= 2;
1997 reclaim_stat
->recent_rotated
[0] /= 2;
2000 if (unlikely(reclaim_stat
->recent_scanned
[1] > file
/ 4)) {
2001 reclaim_stat
->recent_scanned
[1] /= 2;
2002 reclaim_stat
->recent_rotated
[1] /= 2;
2006 * The amount of pressure on anon vs file pages is inversely
2007 * proportional to the fraction of recently scanned pages on
2008 * each list that were recently referenced and in active use.
2010 ap
= anon_prio
* (reclaim_stat
->recent_scanned
[0] + 1);
2011 ap
/= reclaim_stat
->recent_rotated
[0] + 1;
2013 fp
= file_prio
* (reclaim_stat
->recent_scanned
[1] + 1);
2014 fp
/= reclaim_stat
->recent_rotated
[1] + 1;
2015 spin_unlock_irq(&zone
->lru_lock
);
2019 denominator
= ap
+ fp
+ 1;
2021 some_scanned
= false;
2022 /* Only use force_scan on second pass. */
2023 for (pass
= 0; !some_scanned
&& pass
< 2; pass
++) {
2024 for_each_evictable_lru(lru
) {
2025 int file
= is_file_lru(lru
);
2029 size
= get_lru_size(lruvec
, lru
);
2030 scan
= size
>> sc
->priority
;
2032 if (!scan
&& pass
&& force_scan
)
2033 scan
= min(size
, SWAP_CLUSTER_MAX
);
2035 switch (scan_balance
) {
2037 /* Scan lists relative to size */
2041 * Scan types proportional to swappiness and
2042 * their relative recent reclaim efficiency.
2044 scan
= div64_u64(scan
* fraction
[file
],
2049 /* Scan one type exclusively */
2050 if ((scan_balance
== SCAN_FILE
) != file
)
2054 /* Look ma, no brain */
2059 * Skip the second pass and don't force_scan,
2060 * if we found something to scan.
2062 some_scanned
|= !!scan
;
2068 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
2070 static void shrink_lruvec(struct lruvec
*lruvec
, int swappiness
,
2071 struct scan_control
*sc
)
2073 unsigned long nr
[NR_LRU_LISTS
];
2074 unsigned long targets
[NR_LRU_LISTS
];
2075 unsigned long nr_to_scan
;
2077 unsigned long nr_reclaimed
= 0;
2078 unsigned long nr_to_reclaim
= sc
->nr_to_reclaim
;
2079 struct blk_plug plug
;
2082 get_scan_count(lruvec
, swappiness
, sc
, nr
);
2084 /* Record the original scan target for proportional adjustments later */
2085 memcpy(targets
, nr
, sizeof(nr
));
2088 * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal
2089 * event that can occur when there is little memory pressure e.g.
2090 * multiple streaming readers/writers. Hence, we do not abort scanning
2091 * when the requested number of pages are reclaimed when scanning at
2092 * DEF_PRIORITY on the assumption that the fact we are direct
2093 * reclaiming implies that kswapd is not keeping up and it is best to
2094 * do a batch of work at once. For memcg reclaim one check is made to
2095 * abort proportional reclaim if either the file or anon lru has already
2096 * dropped to zero at the first pass.
2098 scan_adjusted
= (global_reclaim(sc
) && !current_is_kswapd() &&
2099 sc
->priority
== DEF_PRIORITY
);
2101 blk_start_plug(&plug
);
2102 while (nr
[LRU_INACTIVE_ANON
] || nr
[LRU_ACTIVE_FILE
] ||
2103 nr
[LRU_INACTIVE_FILE
]) {
2104 unsigned long nr_anon
, nr_file
, percentage
;
2105 unsigned long nr_scanned
;
2107 for_each_evictable_lru(lru
) {
2109 nr_to_scan
= min(nr
[lru
], SWAP_CLUSTER_MAX
);
2110 nr
[lru
] -= nr_to_scan
;
2112 nr_reclaimed
+= shrink_list(lru
, nr_to_scan
,
2117 if (nr_reclaimed
< nr_to_reclaim
|| scan_adjusted
)
2121 * For kswapd and memcg, reclaim at least the number of pages
2122 * requested. Ensure that the anon and file LRUs are scanned
2123 * proportionally what was requested by get_scan_count(). We
2124 * stop reclaiming one LRU and reduce the amount scanning
2125 * proportional to the original scan target.
2127 nr_file
= nr
[LRU_INACTIVE_FILE
] + nr
[LRU_ACTIVE_FILE
];
2128 nr_anon
= nr
[LRU_INACTIVE_ANON
] + nr
[LRU_ACTIVE_ANON
];
2131 * It's just vindictive to attack the larger once the smaller
2132 * has gone to zero. And given the way we stop scanning the
2133 * smaller below, this makes sure that we only make one nudge
2134 * towards proportionality once we've got nr_to_reclaim.
2136 if (!nr_file
|| !nr_anon
)
2139 if (nr_file
> nr_anon
) {
2140 unsigned long scan_target
= targets
[LRU_INACTIVE_ANON
] +
2141 targets
[LRU_ACTIVE_ANON
] + 1;
2143 percentage
= nr_anon
* 100 / scan_target
;
2145 unsigned long scan_target
= targets
[LRU_INACTIVE_FILE
] +
2146 targets
[LRU_ACTIVE_FILE
] + 1;
2148 percentage
= nr_file
* 100 / scan_target
;
2151 /* Stop scanning the smaller of the LRU */
2153 nr
[lru
+ LRU_ACTIVE
] = 0;
2156 * Recalculate the other LRU scan count based on its original
2157 * scan target and the percentage scanning already complete
2159 lru
= (lru
== LRU_FILE
) ? LRU_BASE
: LRU_FILE
;
2160 nr_scanned
= targets
[lru
] - nr
[lru
];
2161 nr
[lru
] = targets
[lru
] * (100 - percentage
) / 100;
2162 nr
[lru
] -= min(nr
[lru
], nr_scanned
);
2165 nr_scanned
= targets
[lru
] - nr
[lru
];
2166 nr
[lru
] = targets
[lru
] * (100 - percentage
) / 100;
2167 nr
[lru
] -= min(nr
[lru
], nr_scanned
);
2169 scan_adjusted
= true;
2171 blk_finish_plug(&plug
);
2172 sc
->nr_reclaimed
+= nr_reclaimed
;
2175 * Even if we did not try to evict anon pages at all, we want to
2176 * rebalance the anon lru active/inactive ratio.
2178 if (inactive_anon_is_low(lruvec
))
2179 shrink_active_list(SWAP_CLUSTER_MAX
, lruvec
,
2180 sc
, LRU_ACTIVE_ANON
);
2182 throttle_vm_writeout(sc
->gfp_mask
);
2185 /* Use reclaim/compaction for costly allocs or under memory pressure */
2186 static bool in_reclaim_compaction(struct scan_control
*sc
)
2188 if (IS_ENABLED(CONFIG_COMPACTION
) && sc
->order
&&
2189 (sc
->order
> PAGE_ALLOC_COSTLY_ORDER
||
2190 sc
->priority
< DEF_PRIORITY
- 2))
2197 * Reclaim/compaction is used for high-order allocation requests. It reclaims
2198 * order-0 pages before compacting the zone. should_continue_reclaim() returns
2199 * true if more pages should be reclaimed such that when the page allocator
2200 * calls try_to_compact_zone() that it will have enough free pages to succeed.
2201 * It will give up earlier than that if there is difficulty reclaiming pages.
2203 static inline bool should_continue_reclaim(struct zone
*zone
,
2204 unsigned long nr_reclaimed
,
2205 unsigned long nr_scanned
,
2206 struct scan_control
*sc
)
2208 unsigned long pages_for_compaction
;
2209 unsigned long inactive_lru_pages
;
2211 /* If not in reclaim/compaction mode, stop */
2212 if (!in_reclaim_compaction(sc
))
2215 /* Consider stopping depending on scan and reclaim activity */
2216 if (sc
->gfp_mask
& __GFP_REPEAT
) {
2218 * For __GFP_REPEAT allocations, stop reclaiming if the
2219 * full LRU list has been scanned and we are still failing
2220 * to reclaim pages. This full LRU scan is potentially
2221 * expensive but a __GFP_REPEAT caller really wants to succeed
2223 if (!nr_reclaimed
&& !nr_scanned
)
2227 * For non-__GFP_REPEAT allocations which can presumably
2228 * fail without consequence, stop if we failed to reclaim
2229 * any pages from the last SWAP_CLUSTER_MAX number of
2230 * pages that were scanned. This will return to the
2231 * caller faster at the risk reclaim/compaction and
2232 * the resulting allocation attempt fails
2239 * If we have not reclaimed enough pages for compaction and the
2240 * inactive lists are large enough, continue reclaiming
2242 pages_for_compaction
= (2UL << sc
->order
);
2243 inactive_lru_pages
= zone_page_state(zone
, NR_INACTIVE_FILE
);
2244 if (get_nr_swap_pages() > 0)
2245 inactive_lru_pages
+= zone_page_state(zone
, NR_INACTIVE_ANON
);
2246 if (sc
->nr_reclaimed
< pages_for_compaction
&&
2247 inactive_lru_pages
> pages_for_compaction
)
2250 /* If compaction would go ahead or the allocation would succeed, stop */
2251 switch (compaction_suitable(zone
, sc
->order
)) {
2252 case COMPACT_PARTIAL
:
2253 case COMPACT_CONTINUE
:
2260 static bool shrink_zone(struct zone
*zone
, struct scan_control
*sc
)
2262 unsigned long nr_reclaimed
, nr_scanned
;
2263 bool reclaimable
= false;
2266 struct mem_cgroup
*root
= sc
->target_mem_cgroup
;
2267 struct mem_cgroup_reclaim_cookie reclaim
= {
2269 .priority
= sc
->priority
,
2271 struct mem_cgroup
*memcg
;
2273 nr_reclaimed
= sc
->nr_reclaimed
;
2274 nr_scanned
= sc
->nr_scanned
;
2276 memcg
= mem_cgroup_iter(root
, NULL
, &reclaim
);
2278 struct lruvec
*lruvec
;
2281 lruvec
= mem_cgroup_zone_lruvec(zone
, memcg
);
2282 swappiness
= mem_cgroup_swappiness(memcg
);
2284 shrink_lruvec(lruvec
, swappiness
, sc
);
2287 * Direct reclaim and kswapd have to scan all memory
2288 * cgroups to fulfill the overall scan target for the
2291 * Limit reclaim, on the other hand, only cares about
2292 * nr_to_reclaim pages to be reclaimed and it will
2293 * retry with decreasing priority if one round over the
2294 * whole hierarchy is not sufficient.
2296 if (!global_reclaim(sc
) &&
2297 sc
->nr_reclaimed
>= sc
->nr_to_reclaim
) {
2298 mem_cgroup_iter_break(root
, memcg
);
2301 memcg
= mem_cgroup_iter(root
, memcg
, &reclaim
);
2304 vmpressure(sc
->gfp_mask
, sc
->target_mem_cgroup
,
2305 sc
->nr_scanned
- nr_scanned
,
2306 sc
->nr_reclaimed
- nr_reclaimed
);
2308 if (sc
->nr_reclaimed
- nr_reclaimed
)
2311 } while (should_continue_reclaim(zone
, sc
->nr_reclaimed
- nr_reclaimed
,
2312 sc
->nr_scanned
- nr_scanned
, sc
));
2318 * Returns true if compaction should go ahead for a high-order request, or
2319 * the high-order allocation would succeed without compaction.
2321 static inline bool compaction_ready(struct zone
*zone
, int order
)
2323 unsigned long balance_gap
, watermark
;
2327 * Compaction takes time to run and there are potentially other
2328 * callers using the pages just freed. Continue reclaiming until
2329 * there is a buffer of free pages available to give compaction
2330 * a reasonable chance of completing and allocating the page
2332 balance_gap
= min(low_wmark_pages(zone
), DIV_ROUND_UP(
2333 zone
->managed_pages
, KSWAPD_ZONE_BALANCE_GAP_RATIO
));
2334 watermark
= high_wmark_pages(zone
) + balance_gap
+ (2UL << order
);
2335 watermark_ok
= zone_watermark_ok_safe(zone
, 0, watermark
, 0, 0);
2338 * If compaction is deferred, reclaim up to a point where
2339 * compaction will have a chance of success when re-enabled
2341 if (compaction_deferred(zone
, order
))
2342 return watermark_ok
;
2345 * If compaction is not ready to start and allocation is not likely
2346 * to succeed without it, then keep reclaiming.
2348 if (compaction_suitable(zone
, order
) == COMPACT_SKIPPED
)
2351 return watermark_ok
;
2355 * This is the direct reclaim path, for page-allocating processes. We only
2356 * try to reclaim pages from zones which will satisfy the caller's allocation
2359 * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
2361 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
2363 * b) The target zone may be at high_wmark_pages(zone) but the lower zones
2364 * must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
2365 * zone defense algorithm.
2367 * If a zone is deemed to be full of pinned pages then just give it a light
2368 * scan then give up on it.
2370 * Returns true if a zone was reclaimable.
2372 static bool shrink_zones(struct zonelist
*zonelist
, struct scan_control
*sc
)
2376 unsigned long nr_soft_reclaimed
;
2377 unsigned long nr_soft_scanned
;
2378 unsigned long lru_pages
= 0;
2379 struct reclaim_state
*reclaim_state
= current
->reclaim_state
;
2381 struct shrink_control shrink
= {
2382 .gfp_mask
= sc
->gfp_mask
,
2384 enum zone_type requested_highidx
= gfp_zone(sc
->gfp_mask
);
2385 bool reclaimable
= false;
2388 * If the number of buffer_heads in the machine exceeds the maximum
2389 * allowed level, force direct reclaim to scan the highmem zone as
2390 * highmem pages could be pinning lowmem pages storing buffer_heads
2392 orig_mask
= sc
->gfp_mask
;
2393 if (buffer_heads_over_limit
)
2394 sc
->gfp_mask
|= __GFP_HIGHMEM
;
2396 nodes_clear(shrink
.nodes_to_scan
);
2398 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
2399 gfp_zone(sc
->gfp_mask
), sc
->nodemask
) {
2400 if (!populated_zone(zone
))
2403 * Take care memory controller reclaiming has small influence
2406 if (global_reclaim(sc
)) {
2407 if (!cpuset_zone_allowed_hardwall(zone
, GFP_KERNEL
))
2410 lru_pages
+= zone_reclaimable_pages(zone
);
2411 node_set(zone_to_nid(zone
), shrink
.nodes_to_scan
);
2413 if (sc
->priority
!= DEF_PRIORITY
&&
2414 !zone_reclaimable(zone
))
2415 continue; /* Let kswapd poll it */
2418 * If we already have plenty of memory free for
2419 * compaction in this zone, don't free any more.
2420 * Even though compaction is invoked for any
2421 * non-zero order, only frequent costly order
2422 * reclamation is disruptive enough to become a
2423 * noticeable problem, like transparent huge
2426 if (IS_ENABLED(CONFIG_COMPACTION
) &&
2427 sc
->order
> PAGE_ALLOC_COSTLY_ORDER
&&
2428 zonelist_zone_idx(z
) <= requested_highidx
&&
2429 compaction_ready(zone
, sc
->order
)) {
2430 sc
->compaction_ready
= true;
2435 * This steals pages from memory cgroups over softlimit
2436 * and returns the number of reclaimed pages and
2437 * scanned pages. This works for global memory pressure
2438 * and balancing, not for a memcg's limit.
2440 nr_soft_scanned
= 0;
2441 nr_soft_reclaimed
= mem_cgroup_soft_limit_reclaim(zone
,
2442 sc
->order
, sc
->gfp_mask
,
2444 sc
->nr_reclaimed
+= nr_soft_reclaimed
;
2445 sc
->nr_scanned
+= nr_soft_scanned
;
2446 if (nr_soft_reclaimed
)
2448 /* need some check for avoid more shrink_zone() */
2451 if (shrink_zone(zone
, sc
))
2454 if (global_reclaim(sc
) &&
2455 !reclaimable
&& zone_reclaimable(zone
))
2460 * Don't shrink slabs when reclaiming memory from over limit cgroups
2461 * but do shrink slab at least once when aborting reclaim for
2462 * compaction to avoid unevenly scanning file/anon LRU pages over slab
2465 if (global_reclaim(sc
)) {
2466 shrink_slab(&shrink
, sc
->nr_scanned
, lru_pages
);
2467 if (reclaim_state
) {
2468 sc
->nr_reclaimed
+= reclaim_state
->reclaimed_slab
;
2469 reclaim_state
->reclaimed_slab
= 0;
2474 * Restore to original mask to avoid the impact on the caller if we
2475 * promoted it to __GFP_HIGHMEM.
2477 sc
->gfp_mask
= orig_mask
;
2483 * This is the main entry point to direct page reclaim.
2485 * If a full scan of the inactive list fails to free enough memory then we
2486 * are "out of memory" and something needs to be killed.
2488 * If the caller is !__GFP_FS then the probability of a failure is reasonably
2489 * high - the zone may be full of dirty or under-writeback pages, which this
2490 * caller can't do much about. We kick the writeback threads and take explicit
2491 * naps in the hope that some of these pages can be written. But if the
2492 * allocating task holds filesystem locks which prevent writeout this might not
2493 * work, and the allocation attempt will fail.
2495 * returns: 0, if no pages reclaimed
2496 * else, the number of pages reclaimed
2498 static unsigned long do_try_to_free_pages(struct zonelist
*zonelist
,
2499 struct scan_control
*sc
)
2501 unsigned long total_scanned
= 0;
2502 unsigned long writeback_threshold
;
2503 bool zones_reclaimable
;
2505 delayacct_freepages_start();
2507 if (global_reclaim(sc
))
2508 count_vm_event(ALLOCSTALL
);
2511 vmpressure_prio(sc
->gfp_mask
, sc
->target_mem_cgroup
,
2514 zones_reclaimable
= shrink_zones(zonelist
, sc
);
2516 total_scanned
+= sc
->nr_scanned
;
2517 if (sc
->nr_reclaimed
>= sc
->nr_to_reclaim
)
2520 if (sc
->compaction_ready
)
2524 * If we're getting trouble reclaiming, start doing
2525 * writepage even in laptop mode.
2527 if (sc
->priority
< DEF_PRIORITY
- 2)
2528 sc
->may_writepage
= 1;
2531 * Try to write back as many pages as we just scanned. This
2532 * tends to cause slow streaming writers to write data to the
2533 * disk smoothly, at the dirtying rate, which is nice. But
2534 * that's undesirable in laptop mode, where we *want* lumpy
2535 * writeout. So in laptop mode, write out the whole world.
2537 writeback_threshold
= sc
->nr_to_reclaim
+ sc
->nr_to_reclaim
/ 2;
2538 if (total_scanned
> writeback_threshold
) {
2539 wakeup_flusher_threads(laptop_mode
? 0 : total_scanned
,
2540 WB_REASON_TRY_TO_FREE_PAGES
);
2541 sc
->may_writepage
= 1;
2543 } while (--sc
->priority
>= 0);
2545 delayacct_freepages_end();
2547 if (sc
->nr_reclaimed
)
2548 return sc
->nr_reclaimed
;
2550 /* Aborted reclaim to try compaction? don't OOM, then */
2551 if (sc
->compaction_ready
)
2554 /* Any of the zones still reclaimable? Don't OOM. */
2555 if (zones_reclaimable
)
2561 static bool pfmemalloc_watermark_ok(pg_data_t
*pgdat
)
2564 unsigned long pfmemalloc_reserve
= 0;
2565 unsigned long free_pages
= 0;
2569 for (i
= 0; i
<= ZONE_NORMAL
; i
++) {
2570 zone
= &pgdat
->node_zones
[i
];
2571 if (!populated_zone(zone
))
2574 pfmemalloc_reserve
+= min_wmark_pages(zone
);
2575 free_pages
+= zone_page_state(zone
, NR_FREE_PAGES
);
2578 /* If there are no reserves (unexpected config) then do not throttle */
2579 if (!pfmemalloc_reserve
)
2582 wmark_ok
= free_pages
> pfmemalloc_reserve
/ 2;
2584 /* kswapd must be awake if processes are being throttled */
2585 if (!wmark_ok
&& waitqueue_active(&pgdat
->kswapd_wait
)) {
2586 pgdat
->classzone_idx
= min(pgdat
->classzone_idx
,
2587 (enum zone_type
)ZONE_NORMAL
);
2588 wake_up_interruptible(&pgdat
->kswapd_wait
);
2595 * Throttle direct reclaimers if backing storage is backed by the network
2596 * and the PFMEMALLOC reserve for the preferred node is getting dangerously
2597 * depleted. kswapd will continue to make progress and wake the processes
2598 * when the low watermark is reached.
2600 * Returns true if a fatal signal was delivered during throttling. If this
2601 * happens, the page allocator should not consider triggering the OOM killer.
2603 static bool throttle_direct_reclaim(gfp_t gfp_mask
, struct zonelist
*zonelist
,
2604 nodemask_t
*nodemask
)
2608 pg_data_t
*pgdat
= NULL
;
2611 * Kernel threads should not be throttled as they may be indirectly
2612 * responsible for cleaning pages necessary for reclaim to make forward
2613 * progress. kjournald for example may enter direct reclaim while
2614 * committing a transaction where throttling it could forcing other
2615 * processes to block on log_wait_commit().
2617 if (current
->flags
& PF_KTHREAD
)
2621 * If a fatal signal is pending, this process should not throttle.
2622 * It should return quickly so it can exit and free its memory
2624 if (fatal_signal_pending(current
))
2628 * Check if the pfmemalloc reserves are ok by finding the first node
2629 * with a usable ZONE_NORMAL or lower zone. The expectation is that
2630 * GFP_KERNEL will be required for allocating network buffers when
2631 * swapping over the network so ZONE_HIGHMEM is unusable.
2633 * Throttling is based on the first usable node and throttled processes
2634 * wait on a queue until kswapd makes progress and wakes them. There
2635 * is an affinity then between processes waking up and where reclaim
2636 * progress has been made assuming the process wakes on the same node.
2637 * More importantly, processes running on remote nodes will not compete
2638 * for remote pfmemalloc reserves and processes on different nodes
2639 * should make reasonable progress.
2641 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
2642 gfp_mask
, nodemask
) {
2643 if (zone_idx(zone
) > ZONE_NORMAL
)
2646 /* Throttle based on the first usable node */
2647 pgdat
= zone
->zone_pgdat
;
2648 if (pfmemalloc_watermark_ok(pgdat
))
2653 /* If no zone was usable by the allocation flags then do not throttle */
2657 /* Account for the throttling */
2658 count_vm_event(PGSCAN_DIRECT_THROTTLE
);
2661 * If the caller cannot enter the filesystem, it's possible that it
2662 * is due to the caller holding an FS lock or performing a journal
2663 * transaction in the case of a filesystem like ext[3|4]. In this case,
2664 * it is not safe to block on pfmemalloc_wait as kswapd could be
2665 * blocked waiting on the same lock. Instead, throttle for up to a
2666 * second before continuing.
2668 if (!(gfp_mask
& __GFP_FS
)) {
2669 wait_event_interruptible_timeout(pgdat
->pfmemalloc_wait
,
2670 pfmemalloc_watermark_ok(pgdat
), HZ
);
2675 /* Throttle until kswapd wakes the process */
2676 wait_event_killable(zone
->zone_pgdat
->pfmemalloc_wait
,
2677 pfmemalloc_watermark_ok(pgdat
));
2680 if (fatal_signal_pending(current
))
2687 unsigned long try_to_free_pages(struct zonelist
*zonelist
, int order
,
2688 gfp_t gfp_mask
, nodemask_t
*nodemask
)
2690 unsigned long nr_reclaimed
;
2691 struct scan_control sc
= {
2692 .nr_to_reclaim
= SWAP_CLUSTER_MAX
,
2693 .gfp_mask
= (gfp_mask
= memalloc_noio_flags(gfp_mask
)),
2695 .nodemask
= nodemask
,
2696 .priority
= DEF_PRIORITY
,
2697 .may_writepage
= !laptop_mode
,
2703 * Do not enter reclaim if fatal signal was delivered while throttled.
2704 * 1 is returned so that the page allocator does not OOM kill at this
2707 if (throttle_direct_reclaim(gfp_mask
, zonelist
, nodemask
))
2710 trace_mm_vmscan_direct_reclaim_begin(order
,
2714 nr_reclaimed
= do_try_to_free_pages(zonelist
, &sc
);
2716 trace_mm_vmscan_direct_reclaim_end(nr_reclaimed
);
2718 return nr_reclaimed
;
2723 unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup
*memcg
,
2724 gfp_t gfp_mask
, bool noswap
,
2726 unsigned long *nr_scanned
)
2728 struct scan_control sc
= {
2729 .nr_to_reclaim
= SWAP_CLUSTER_MAX
,
2730 .target_mem_cgroup
= memcg
,
2731 .may_writepage
= !laptop_mode
,
2733 .may_swap
= !noswap
,
2735 struct lruvec
*lruvec
= mem_cgroup_zone_lruvec(zone
, memcg
);
2736 int swappiness
= mem_cgroup_swappiness(memcg
);
2738 sc
.gfp_mask
= (gfp_mask
& GFP_RECLAIM_MASK
) |
2739 (GFP_HIGHUSER_MOVABLE
& ~GFP_RECLAIM_MASK
);
2741 trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc
.order
,
2746 * NOTE: Although we can get the priority field, using it
2747 * here is not a good idea, since it limits the pages we can scan.
2748 * if we don't reclaim here, the shrink_zone from balance_pgdat
2749 * will pick up pages from other mem cgroup's as well. We hack
2750 * the priority and make it zero.
2752 shrink_lruvec(lruvec
, swappiness
, &sc
);
2754 trace_mm_vmscan_memcg_softlimit_reclaim_end(sc
.nr_reclaimed
);
2756 *nr_scanned
= sc
.nr_scanned
;
2757 return sc
.nr_reclaimed
;
2760 unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup
*memcg
,
2761 unsigned long nr_pages
,
2765 struct zonelist
*zonelist
;
2766 unsigned long nr_reclaimed
;
2768 struct scan_control sc
= {
2769 .nr_to_reclaim
= max(nr_pages
, SWAP_CLUSTER_MAX
),
2770 .gfp_mask
= (gfp_mask
& GFP_RECLAIM_MASK
) |
2771 (GFP_HIGHUSER_MOVABLE
& ~GFP_RECLAIM_MASK
),
2772 .target_mem_cgroup
= memcg
,
2773 .priority
= DEF_PRIORITY
,
2774 .may_writepage
= !laptop_mode
,
2776 .may_swap
= may_swap
,
2780 * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't
2781 * take care of from where we get pages. So the node where we start the
2782 * scan does not need to be the current node.
2784 nid
= mem_cgroup_select_victim_node(memcg
);
2786 zonelist
= NODE_DATA(nid
)->node_zonelists
;
2788 trace_mm_vmscan_memcg_reclaim_begin(0,
2792 nr_reclaimed
= do_try_to_free_pages(zonelist
, &sc
);
2794 trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed
);
2796 return nr_reclaimed
;
2800 static void age_active_anon(struct zone
*zone
, struct scan_control
*sc
)
2802 struct mem_cgroup
*memcg
;
2804 if (!total_swap_pages
)
2807 memcg
= mem_cgroup_iter(NULL
, NULL
, NULL
);
2809 struct lruvec
*lruvec
= mem_cgroup_zone_lruvec(zone
, memcg
);
2811 if (inactive_anon_is_low(lruvec
))
2812 shrink_active_list(SWAP_CLUSTER_MAX
, lruvec
,
2813 sc
, LRU_ACTIVE_ANON
);
2815 memcg
= mem_cgroup_iter(NULL
, memcg
, NULL
);
2819 static bool zone_balanced(struct zone
*zone
, int order
,
2820 unsigned long balance_gap
, int classzone_idx
)
2822 if (!zone_watermark_ok_safe(zone
, order
, high_wmark_pages(zone
) +
2823 balance_gap
, classzone_idx
, 0))
2826 if (IS_ENABLED(CONFIG_COMPACTION
) && order
&&
2827 compaction_suitable(zone
, order
) == COMPACT_SKIPPED
)
2834 * pgdat_balanced() is used when checking if a node is balanced.
2836 * For order-0, all zones must be balanced!
2838 * For high-order allocations only zones that meet watermarks and are in a
2839 * zone allowed by the callers classzone_idx are added to balanced_pages. The
2840 * total of balanced pages must be at least 25% of the zones allowed by
2841 * classzone_idx for the node to be considered balanced. Forcing all zones to
2842 * be balanced for high orders can cause excessive reclaim when there are
2844 * The choice of 25% is due to
2845 * o a 16M DMA zone that is balanced will not balance a zone on any
2846 * reasonable sized machine
2847 * o On all other machines, the top zone must be at least a reasonable
2848 * percentage of the middle zones. For example, on 32-bit x86, highmem
2849 * would need to be at least 256M for it to be balance a whole node.
2850 * Similarly, on x86-64 the Normal zone would need to be at least 1G
2851 * to balance a node on its own. These seemed like reasonable ratios.
2853 static bool pgdat_balanced(pg_data_t
*pgdat
, int order
, int classzone_idx
)
2855 unsigned long managed_pages
= 0;
2856 unsigned long balanced_pages
= 0;
2859 /* Check the watermark levels */
2860 for (i
= 0; i
<= classzone_idx
; i
++) {
2861 struct zone
*zone
= pgdat
->node_zones
+ i
;
2863 if (!populated_zone(zone
))
2866 managed_pages
+= zone
->managed_pages
;
2869 * A special case here:
2871 * balance_pgdat() skips over all_unreclaimable after
2872 * DEF_PRIORITY. Effectively, it considers them balanced so
2873 * they must be considered balanced here as well!
2875 if (!zone_reclaimable(zone
)) {
2876 balanced_pages
+= zone
->managed_pages
;
2880 if (zone_balanced(zone
, order
, 0, i
))
2881 balanced_pages
+= zone
->managed_pages
;
2887 return balanced_pages
>= (managed_pages
>> 2);
2893 * Prepare kswapd for sleeping. This verifies that there are no processes
2894 * waiting in throttle_direct_reclaim() and that watermarks have been met.
2896 * Returns true if kswapd is ready to sleep
2898 static bool prepare_kswapd_sleep(pg_data_t
*pgdat
, int order
, long remaining
,
2901 /* If a direct reclaimer woke kswapd within HZ/10, it's premature */
2906 * There is a potential race between when kswapd checks its watermarks
2907 * and a process gets throttled. There is also a potential race if
2908 * processes get throttled, kswapd wakes, a large process exits therby
2909 * balancing the zones that causes kswapd to miss a wakeup. If kswapd
2910 * is going to sleep, no process should be sleeping on pfmemalloc_wait
2911 * so wake them now if necessary. If necessary, processes will wake
2912 * kswapd and get throttled again
2914 if (waitqueue_active(&pgdat
->pfmemalloc_wait
)) {
2915 wake_up(&pgdat
->pfmemalloc_wait
);
2919 return pgdat_balanced(pgdat
, order
, classzone_idx
);
2923 * kswapd shrinks the zone by the number of pages required to reach
2924 * the high watermark.
2926 * Returns true if kswapd scanned at least the requested number of pages to
2927 * reclaim or if the lack of progress was due to pages under writeback.
2928 * This is used to determine if the scanning priority needs to be raised.
2930 static bool kswapd_shrink_zone(struct zone
*zone
,
2932 struct scan_control
*sc
,
2933 unsigned long lru_pages
,
2934 unsigned long *nr_attempted
)
2936 int testorder
= sc
->order
;
2937 unsigned long balance_gap
;
2938 struct reclaim_state
*reclaim_state
= current
->reclaim_state
;
2939 struct shrink_control shrink
= {
2940 .gfp_mask
= sc
->gfp_mask
,
2942 bool lowmem_pressure
;
2944 /* Reclaim above the high watermark. */
2945 sc
->nr_to_reclaim
= max(SWAP_CLUSTER_MAX
, high_wmark_pages(zone
));
2948 * Kswapd reclaims only single pages with compaction enabled. Trying
2949 * too hard to reclaim until contiguous free pages have become
2950 * available can hurt performance by evicting too much useful data
2951 * from memory. Do not reclaim more than needed for compaction.
2953 if (IS_ENABLED(CONFIG_COMPACTION
) && sc
->order
&&
2954 compaction_suitable(zone
, sc
->order
) !=
2959 * We put equal pressure on every zone, unless one zone has way too
2960 * many pages free already. The "too many pages" is defined as the
2961 * high wmark plus a "gap" where the gap is either the low
2962 * watermark or 1% of the zone, whichever is smaller.
2964 balance_gap
= min(low_wmark_pages(zone
), DIV_ROUND_UP(
2965 zone
->managed_pages
, KSWAPD_ZONE_BALANCE_GAP_RATIO
));
2968 * If there is no low memory pressure or the zone is balanced then no
2969 * reclaim is necessary
2971 lowmem_pressure
= (buffer_heads_over_limit
&& is_highmem(zone
));
2972 if (!lowmem_pressure
&& zone_balanced(zone
, testorder
,
2973 balance_gap
, classzone_idx
))
2976 shrink_zone(zone
, sc
);
2977 nodes_clear(shrink
.nodes_to_scan
);
2978 node_set(zone_to_nid(zone
), shrink
.nodes_to_scan
);
2980 reclaim_state
->reclaimed_slab
= 0;
2981 shrink_slab(&shrink
, sc
->nr_scanned
, lru_pages
);
2982 sc
->nr_reclaimed
+= reclaim_state
->reclaimed_slab
;
2984 /* Account for the number of pages attempted to reclaim */
2985 *nr_attempted
+= sc
->nr_to_reclaim
;
2987 clear_bit(ZONE_WRITEBACK
, &zone
->flags
);
2990 * If a zone reaches its high watermark, consider it to be no longer
2991 * congested. It's possible there are dirty pages backed by congested
2992 * BDIs but as pressure is relieved, speculatively avoid congestion
2995 if (zone_reclaimable(zone
) &&
2996 zone_balanced(zone
, testorder
, 0, classzone_idx
)) {
2997 clear_bit(ZONE_CONGESTED
, &zone
->flags
);
2998 clear_bit(ZONE_DIRTY
, &zone
->flags
);
3001 return sc
->nr_scanned
>= sc
->nr_to_reclaim
;
3005 * For kswapd, balance_pgdat() will work across all this node's zones until
3006 * they are all at high_wmark_pages(zone).
3008 * Returns the final order kswapd was reclaiming at
3010 * There is special handling here for zones which are full of pinned pages.
3011 * This can happen if the pages are all mlocked, or if they are all used by
3012 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
3013 * What we do is to detect the case where all pages in the zone have been
3014 * scanned twice and there has been zero successful reclaim. Mark the zone as
3015 * dead and from now on, only perform a short scan. Basically we're polling
3016 * the zone for when the problem goes away.
3018 * kswapd scans the zones in the highmem->normal->dma direction. It skips
3019 * zones which have free_pages > high_wmark_pages(zone), but once a zone is
3020 * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the
3021 * lower zones regardless of the number of free pages in the lower zones. This
3022 * interoperates with the page allocator fallback scheme to ensure that aging
3023 * of pages is balanced across the zones.
3025 static unsigned long balance_pgdat(pg_data_t
*pgdat
, int order
,
3029 int end_zone
= 0; /* Inclusive. 0 = ZONE_DMA */
3030 unsigned long nr_soft_reclaimed
;
3031 unsigned long nr_soft_scanned
;
3032 struct scan_control sc
= {
3033 .gfp_mask
= GFP_KERNEL
,
3035 .priority
= DEF_PRIORITY
,
3036 .may_writepage
= !laptop_mode
,
3040 count_vm_event(PAGEOUTRUN
);
3043 unsigned long lru_pages
= 0;
3044 unsigned long nr_attempted
= 0;
3045 bool raise_priority
= true;
3046 bool pgdat_needs_compaction
= (order
> 0);
3048 sc
.nr_reclaimed
= 0;
3051 * Scan in the highmem->dma direction for the highest
3052 * zone which needs scanning
3054 for (i
= pgdat
->nr_zones
- 1; i
>= 0; i
--) {
3055 struct zone
*zone
= pgdat
->node_zones
+ i
;
3057 if (!populated_zone(zone
))
3060 if (sc
.priority
!= DEF_PRIORITY
&&
3061 !zone_reclaimable(zone
))
3065 * Do some background aging of the anon list, to give
3066 * pages a chance to be referenced before reclaiming.
3068 age_active_anon(zone
, &sc
);
3071 * If the number of buffer_heads in the machine
3072 * exceeds the maximum allowed level and this node
3073 * has a highmem zone, force kswapd to reclaim from
3074 * it to relieve lowmem pressure.
3076 if (buffer_heads_over_limit
&& is_highmem_idx(i
)) {
3081 if (!zone_balanced(zone
, order
, 0, 0)) {
3086 * If balanced, clear the dirty and congested
3089 clear_bit(ZONE_CONGESTED
, &zone
->flags
);
3090 clear_bit(ZONE_DIRTY
, &zone
->flags
);
3097 for (i
= 0; i
<= end_zone
; i
++) {
3098 struct zone
*zone
= pgdat
->node_zones
+ i
;
3100 if (!populated_zone(zone
))
3103 lru_pages
+= zone_reclaimable_pages(zone
);
3106 * If any zone is currently balanced then kswapd will
3107 * not call compaction as it is expected that the
3108 * necessary pages are already available.
3110 if (pgdat_needs_compaction
&&
3111 zone_watermark_ok(zone
, order
,
3112 low_wmark_pages(zone
),
3114 pgdat_needs_compaction
= false;
3118 * If we're getting trouble reclaiming, start doing writepage
3119 * even in laptop mode.
3121 if (sc
.priority
< DEF_PRIORITY
- 2)
3122 sc
.may_writepage
= 1;
3125 * Now scan the zone in the dma->highmem direction, stopping
3126 * at the last zone which needs scanning.
3128 * We do this because the page allocator works in the opposite
3129 * direction. This prevents the page allocator from allocating
3130 * pages behind kswapd's direction of progress, which would
3131 * cause too much scanning of the lower zones.
3133 for (i
= 0; i
<= end_zone
; i
++) {
3134 struct zone
*zone
= pgdat
->node_zones
+ i
;
3136 if (!populated_zone(zone
))
3139 if (sc
.priority
!= DEF_PRIORITY
&&
3140 !zone_reclaimable(zone
))
3145 nr_soft_scanned
= 0;
3147 * Call soft limit reclaim before calling shrink_zone.
3149 nr_soft_reclaimed
= mem_cgroup_soft_limit_reclaim(zone
,
3152 sc
.nr_reclaimed
+= nr_soft_reclaimed
;
3155 * There should be no need to raise the scanning
3156 * priority if enough pages are already being scanned
3157 * that that high watermark would be met at 100%
3160 if (kswapd_shrink_zone(zone
, end_zone
, &sc
,
3161 lru_pages
, &nr_attempted
))
3162 raise_priority
= false;
3166 * If the low watermark is met there is no need for processes
3167 * to be throttled on pfmemalloc_wait as they should not be
3168 * able to safely make forward progress. Wake them
3170 if (waitqueue_active(&pgdat
->pfmemalloc_wait
) &&
3171 pfmemalloc_watermark_ok(pgdat
))
3172 wake_up(&pgdat
->pfmemalloc_wait
);
3175 * Fragmentation may mean that the system cannot be rebalanced
3176 * for high-order allocations in all zones. If twice the
3177 * allocation size has been reclaimed and the zones are still
3178 * not balanced then recheck the watermarks at order-0 to
3179 * prevent kswapd reclaiming excessively. Assume that a
3180 * process requested a high-order can direct reclaim/compact.
3182 if (order
&& sc
.nr_reclaimed
>= 2UL << order
)
3183 order
= sc
.order
= 0;
3185 /* Check if kswapd should be suspending */
3186 if (try_to_freeze() || kthread_should_stop())
3190 * Compact if necessary and kswapd is reclaiming at least the
3191 * high watermark number of pages as requsted
3193 if (pgdat_needs_compaction
&& sc
.nr_reclaimed
> nr_attempted
)
3194 compact_pgdat(pgdat
, order
);
3197 * Raise priority if scanning rate is too low or there was no
3198 * progress in reclaiming pages
3200 if (raise_priority
|| !sc
.nr_reclaimed
)
3202 } while (sc
.priority
>= 1 &&
3203 !pgdat_balanced(pgdat
, order
, *classzone_idx
));
3207 * Return the order we were reclaiming at so prepare_kswapd_sleep()
3208 * makes a decision on the order we were last reclaiming at. However,
3209 * if another caller entered the allocator slow path while kswapd
3210 * was awake, order will remain at the higher level
3212 *classzone_idx
= end_zone
;
3216 static void kswapd_try_to_sleep(pg_data_t
*pgdat
, int order
, int classzone_idx
)
3221 if (freezing(current
) || kthread_should_stop())
3224 prepare_to_wait(&pgdat
->kswapd_wait
, &wait
, TASK_INTERRUPTIBLE
);
3226 /* Try to sleep for a short interval */
3227 if (prepare_kswapd_sleep(pgdat
, order
, remaining
, classzone_idx
)) {
3228 remaining
= schedule_timeout(HZ
/10);
3229 finish_wait(&pgdat
->kswapd_wait
, &wait
);
3230 prepare_to_wait(&pgdat
->kswapd_wait
, &wait
, TASK_INTERRUPTIBLE
);
3234 * After a short sleep, check if it was a premature sleep. If not, then
3235 * go fully to sleep until explicitly woken up.
3237 if (prepare_kswapd_sleep(pgdat
, order
, remaining
, classzone_idx
)) {
3238 trace_mm_vmscan_kswapd_sleep(pgdat
->node_id
);
3241 * vmstat counters are not perfectly accurate and the estimated
3242 * value for counters such as NR_FREE_PAGES can deviate from the
3243 * true value by nr_online_cpus * threshold. To avoid the zone
3244 * watermarks being breached while under pressure, we reduce the
3245 * per-cpu vmstat threshold while kswapd is awake and restore
3246 * them before going back to sleep.
3248 set_pgdat_percpu_threshold(pgdat
, calculate_normal_threshold
);
3251 * Compaction records what page blocks it recently failed to
3252 * isolate pages from and skips them in the future scanning.
3253 * When kswapd is going to sleep, it is reasonable to assume
3254 * that pages and compaction may succeed so reset the cache.
3256 reset_isolation_suitable(pgdat
);
3258 if (!kthread_should_stop())
3261 set_pgdat_percpu_threshold(pgdat
, calculate_pressure_threshold
);
3264 count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY
);
3266 count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY
);
3268 finish_wait(&pgdat
->kswapd_wait
, &wait
);
3272 * The background pageout daemon, started as a kernel thread
3273 * from the init process.
3275 * This basically trickles out pages so that we have _some_
3276 * free memory available even if there is no other activity
3277 * that frees anything up. This is needed for things like routing
3278 * etc, where we otherwise might have all activity going on in
3279 * asynchronous contexts that cannot page things out.
3281 * If there are applications that are active memory-allocators
3282 * (most normal use), this basically shouldn't matter.
3284 static int kswapd(void *p
)
3286 unsigned long order
, new_order
;
3287 unsigned balanced_order
;
3288 int classzone_idx
, new_classzone_idx
;
3289 int balanced_classzone_idx
;
3290 pg_data_t
*pgdat
= (pg_data_t
*)p
;
3291 struct task_struct
*tsk
= current
;
3293 struct reclaim_state reclaim_state
= {
3294 .reclaimed_slab
= 0,
3296 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
3298 lockdep_set_current_reclaim_state(GFP_KERNEL
);
3300 if (!cpumask_empty(cpumask
))
3301 set_cpus_allowed_ptr(tsk
, cpumask
);
3302 current
->reclaim_state
= &reclaim_state
;
3305 * Tell the memory management that we're a "memory allocator",
3306 * and that if we need more memory we should get access to it
3307 * regardless (see "__alloc_pages()"). "kswapd" should
3308 * never get caught in the normal page freeing logic.
3310 * (Kswapd normally doesn't need memory anyway, but sometimes
3311 * you need a small amount of memory in order to be able to
3312 * page out something else, and this flag essentially protects
3313 * us from recursively trying to free more memory as we're
3314 * trying to free the first piece of memory in the first place).
3316 tsk
->flags
|= PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
;
3319 order
= new_order
= 0;
3321 classzone_idx
= new_classzone_idx
= pgdat
->nr_zones
- 1;
3322 balanced_classzone_idx
= classzone_idx
;
3327 * If the last balance_pgdat was unsuccessful it's unlikely a
3328 * new request of a similar or harder type will succeed soon
3329 * so consider going to sleep on the basis we reclaimed at
3331 if (balanced_classzone_idx
>= new_classzone_idx
&&
3332 balanced_order
== new_order
) {
3333 new_order
= pgdat
->kswapd_max_order
;
3334 new_classzone_idx
= pgdat
->classzone_idx
;
3335 pgdat
->kswapd_max_order
= 0;
3336 pgdat
->classzone_idx
= pgdat
->nr_zones
- 1;
3339 if (order
< new_order
|| classzone_idx
> new_classzone_idx
) {
3341 * Don't sleep if someone wants a larger 'order'
3342 * allocation or has tigher zone constraints
3345 classzone_idx
= new_classzone_idx
;
3347 kswapd_try_to_sleep(pgdat
, balanced_order
,
3348 balanced_classzone_idx
);
3349 order
= pgdat
->kswapd_max_order
;
3350 classzone_idx
= pgdat
->classzone_idx
;
3352 new_classzone_idx
= classzone_idx
;
3353 pgdat
->kswapd_max_order
= 0;
3354 pgdat
->classzone_idx
= pgdat
->nr_zones
- 1;
3357 ret
= try_to_freeze();
3358 if (kthread_should_stop())
3362 * We can speed up thawing tasks if we don't call balance_pgdat
3363 * after returning from the refrigerator
3366 trace_mm_vmscan_kswapd_wake(pgdat
->node_id
, order
);
3367 balanced_classzone_idx
= classzone_idx
;
3368 balanced_order
= balance_pgdat(pgdat
, order
,
3369 &balanced_classzone_idx
);
3373 tsk
->flags
&= ~(PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
);
3374 current
->reclaim_state
= NULL
;
3375 lockdep_clear_current_reclaim_state();
3381 * A zone is low on free memory, so wake its kswapd task to service it.
3383 void wakeup_kswapd(struct zone
*zone
, int order
, enum zone_type classzone_idx
)
3387 if (!populated_zone(zone
))
3390 if (!cpuset_zone_allowed_hardwall(zone
, GFP_KERNEL
))
3392 pgdat
= zone
->zone_pgdat
;
3393 if (pgdat
->kswapd_max_order
< order
) {
3394 pgdat
->kswapd_max_order
= order
;
3395 pgdat
->classzone_idx
= min(pgdat
->classzone_idx
, classzone_idx
);
3397 if (!waitqueue_active(&pgdat
->kswapd_wait
))
3399 if (zone_balanced(zone
, order
, 0, 0))
3402 trace_mm_vmscan_wakeup_kswapd(pgdat
->node_id
, zone_idx(zone
), order
);
3403 wake_up_interruptible(&pgdat
->kswapd_wait
);
3406 #ifdef CONFIG_HIBERNATION
3408 * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of
3411 * Rather than trying to age LRUs the aim is to preserve the overall
3412 * LRU order by reclaiming preferentially
3413 * inactive > active > active referenced > active mapped
3415 unsigned long shrink_all_memory(unsigned long nr_to_reclaim
)
3417 struct reclaim_state reclaim_state
;
3418 struct scan_control sc
= {
3419 .nr_to_reclaim
= nr_to_reclaim
,
3420 .gfp_mask
= GFP_HIGHUSER_MOVABLE
,
3421 .priority
= DEF_PRIORITY
,
3425 .hibernation_mode
= 1,
3427 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), sc
.gfp_mask
);
3428 struct task_struct
*p
= current
;
3429 unsigned long nr_reclaimed
;
3431 p
->flags
|= PF_MEMALLOC
;
3432 lockdep_set_current_reclaim_state(sc
.gfp_mask
);
3433 reclaim_state
.reclaimed_slab
= 0;
3434 p
->reclaim_state
= &reclaim_state
;
3436 nr_reclaimed
= do_try_to_free_pages(zonelist
, &sc
);
3438 p
->reclaim_state
= NULL
;
3439 lockdep_clear_current_reclaim_state();
3440 p
->flags
&= ~PF_MEMALLOC
;
3442 return nr_reclaimed
;
3444 #endif /* CONFIG_HIBERNATION */
3446 /* It's optimal to keep kswapds on the same CPUs as their memory, but
3447 not required for correctness. So if the last cpu in a node goes
3448 away, we get changed to run anywhere: as the first one comes back,
3449 restore their cpu bindings. */
3450 static int cpu_callback(struct notifier_block
*nfb
, unsigned long action
,
3455 if (action
== CPU_ONLINE
|| action
== CPU_ONLINE_FROZEN
) {
3456 for_each_node_state(nid
, N_MEMORY
) {
3457 pg_data_t
*pgdat
= NODE_DATA(nid
);
3458 const struct cpumask
*mask
;
3460 mask
= cpumask_of_node(pgdat
->node_id
);
3462 if (cpumask_any_and(cpu_online_mask
, mask
) < nr_cpu_ids
)
3463 /* One of our CPUs online: restore mask */
3464 set_cpus_allowed_ptr(pgdat
->kswapd
, mask
);
3471 * This kswapd start function will be called by init and node-hot-add.
3472 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
3474 int kswapd_run(int nid
)
3476 pg_data_t
*pgdat
= NODE_DATA(nid
);
3482 pgdat
->kswapd
= kthread_run(kswapd
, pgdat
, "kswapd%d", nid
);
3483 if (IS_ERR(pgdat
->kswapd
)) {
3484 /* failure at boot is fatal */
3485 BUG_ON(system_state
== SYSTEM_BOOTING
);
3486 pr_err("Failed to start kswapd on node %d\n", nid
);
3487 ret
= PTR_ERR(pgdat
->kswapd
);
3488 pgdat
->kswapd
= NULL
;
3494 * Called by memory hotplug when all memory in a node is offlined. Caller must
3495 * hold mem_hotplug_begin/end().
3497 void kswapd_stop(int nid
)
3499 struct task_struct
*kswapd
= NODE_DATA(nid
)->kswapd
;
3502 kthread_stop(kswapd
);
3503 NODE_DATA(nid
)->kswapd
= NULL
;
3507 static int __init
kswapd_init(void)
3512 for_each_node_state(nid
, N_MEMORY
)
3514 hotcpu_notifier(cpu_callback
, 0);
3518 module_init(kswapd_init
)
3524 * If non-zero call zone_reclaim when the number of free pages falls below
3527 int zone_reclaim_mode __read_mostly
;
3529 #define RECLAIM_OFF 0
3530 #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
3531 #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
3532 #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
3535 * Priority for ZONE_RECLAIM. This determines the fraction of pages
3536 * of a node considered for each zone_reclaim. 4 scans 1/16th of
3539 #define ZONE_RECLAIM_PRIORITY 4
3542 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
3545 int sysctl_min_unmapped_ratio
= 1;
3548 * If the number of slab pages in a zone grows beyond this percentage then
3549 * slab reclaim needs to occur.
3551 int sysctl_min_slab_ratio
= 5;
3553 static inline unsigned long zone_unmapped_file_pages(struct zone
*zone
)
3555 unsigned long file_mapped
= zone_page_state(zone
, NR_FILE_MAPPED
);
3556 unsigned long file_lru
= zone_page_state(zone
, NR_INACTIVE_FILE
) +
3557 zone_page_state(zone
, NR_ACTIVE_FILE
);
3560 * It's possible for there to be more file mapped pages than
3561 * accounted for by the pages on the file LRU lists because
3562 * tmpfs pages accounted for as ANON can also be FILE_MAPPED
3564 return (file_lru
> file_mapped
) ? (file_lru
- file_mapped
) : 0;
3567 /* Work out how many page cache pages we can reclaim in this reclaim_mode */
3568 static long zone_pagecache_reclaimable(struct zone
*zone
)
3570 long nr_pagecache_reclaimable
;
3574 * If RECLAIM_SWAP is set, then all file pages are considered
3575 * potentially reclaimable. Otherwise, we have to worry about
3576 * pages like swapcache and zone_unmapped_file_pages() provides
3579 if (zone_reclaim_mode
& RECLAIM_SWAP
)
3580 nr_pagecache_reclaimable
= zone_page_state(zone
, NR_FILE_PAGES
);
3582 nr_pagecache_reclaimable
= zone_unmapped_file_pages(zone
);
3584 /* If we can't clean pages, remove dirty pages from consideration */
3585 if (!(zone_reclaim_mode
& RECLAIM_WRITE
))
3586 delta
+= zone_page_state(zone
, NR_FILE_DIRTY
);
3588 /* Watch for any possible underflows due to delta */
3589 if (unlikely(delta
> nr_pagecache_reclaimable
))
3590 delta
= nr_pagecache_reclaimable
;
3592 return nr_pagecache_reclaimable
- delta
;
3596 * Try to free up some pages from this zone through reclaim.
3598 static int __zone_reclaim(struct zone
*zone
, gfp_t gfp_mask
, unsigned int order
)
3600 /* Minimum pages needed in order to stay on node */
3601 const unsigned long nr_pages
= 1 << order
;
3602 struct task_struct
*p
= current
;
3603 struct reclaim_state reclaim_state
;
3604 struct scan_control sc
= {
3605 .nr_to_reclaim
= max(nr_pages
, SWAP_CLUSTER_MAX
),
3606 .gfp_mask
= (gfp_mask
= memalloc_noio_flags(gfp_mask
)),
3608 .priority
= ZONE_RECLAIM_PRIORITY
,
3609 .may_writepage
= !!(zone_reclaim_mode
& RECLAIM_WRITE
),
3610 .may_unmap
= !!(zone_reclaim_mode
& RECLAIM_SWAP
),
3613 struct shrink_control shrink
= {
3614 .gfp_mask
= sc
.gfp_mask
,
3616 unsigned long nr_slab_pages0
, nr_slab_pages1
;
3620 * We need to be able to allocate from the reserves for RECLAIM_SWAP
3621 * and we also need to be able to write out pages for RECLAIM_WRITE
3624 p
->flags
|= PF_MEMALLOC
| PF_SWAPWRITE
;
3625 lockdep_set_current_reclaim_state(gfp_mask
);
3626 reclaim_state
.reclaimed_slab
= 0;
3627 p
->reclaim_state
= &reclaim_state
;
3629 if (zone_pagecache_reclaimable(zone
) > zone
->min_unmapped_pages
) {
3631 * Free memory by calling shrink zone with increasing
3632 * priorities until we have enough memory freed.
3635 shrink_zone(zone
, &sc
);
3636 } while (sc
.nr_reclaimed
< nr_pages
&& --sc
.priority
>= 0);
3639 nr_slab_pages0
= zone_page_state(zone
, NR_SLAB_RECLAIMABLE
);
3640 if (nr_slab_pages0
> zone
->min_slab_pages
) {
3642 * shrink_slab() does not currently allow us to determine how
3643 * many pages were freed in this zone. So we take the current
3644 * number of slab pages and shake the slab until it is reduced
3645 * by the same nr_pages that we used for reclaiming unmapped
3648 nodes_clear(shrink
.nodes_to_scan
);
3649 node_set(zone_to_nid(zone
), shrink
.nodes_to_scan
);
3651 unsigned long lru_pages
= zone_reclaimable_pages(zone
);
3653 /* No reclaimable slab or very low memory pressure */
3654 if (!shrink_slab(&shrink
, sc
.nr_scanned
, lru_pages
))
3657 /* Freed enough memory */
3658 nr_slab_pages1
= zone_page_state(zone
,
3659 NR_SLAB_RECLAIMABLE
);
3660 if (nr_slab_pages1
+ nr_pages
<= nr_slab_pages0
)
3665 * Update nr_reclaimed by the number of slab pages we
3666 * reclaimed from this zone.
3668 nr_slab_pages1
= zone_page_state(zone
, NR_SLAB_RECLAIMABLE
);
3669 if (nr_slab_pages1
< nr_slab_pages0
)
3670 sc
.nr_reclaimed
+= nr_slab_pages0
- nr_slab_pages1
;
3673 p
->reclaim_state
= NULL
;
3674 current
->flags
&= ~(PF_MEMALLOC
| PF_SWAPWRITE
);
3675 lockdep_clear_current_reclaim_state();
3676 return sc
.nr_reclaimed
>= nr_pages
;
3679 int zone_reclaim(struct zone
*zone
, gfp_t gfp_mask
, unsigned int order
)
3685 * Zone reclaim reclaims unmapped file backed pages and
3686 * slab pages if we are over the defined limits.
3688 * A small portion of unmapped file backed pages is needed for
3689 * file I/O otherwise pages read by file I/O will be immediately
3690 * thrown out if the zone is overallocated. So we do not reclaim
3691 * if less than a specified percentage of the zone is used by
3692 * unmapped file backed pages.
3694 if (zone_pagecache_reclaimable(zone
) <= zone
->min_unmapped_pages
&&
3695 zone_page_state(zone
, NR_SLAB_RECLAIMABLE
) <= zone
->min_slab_pages
)
3696 return ZONE_RECLAIM_FULL
;
3698 if (!zone_reclaimable(zone
))
3699 return ZONE_RECLAIM_FULL
;
3702 * Do not scan if the allocation should not be delayed.
3704 if (!(gfp_mask
& __GFP_WAIT
) || (current
->flags
& PF_MEMALLOC
))
3705 return ZONE_RECLAIM_NOSCAN
;
3708 * Only run zone reclaim on the local zone or on zones that do not
3709 * have associated processors. This will favor the local processor
3710 * over remote processors and spread off node memory allocations
3711 * as wide as possible.
3713 node_id
= zone_to_nid(zone
);
3714 if (node_state(node_id
, N_CPU
) && node_id
!= numa_node_id())
3715 return ZONE_RECLAIM_NOSCAN
;
3717 if (test_and_set_bit(ZONE_RECLAIM_LOCKED
, &zone
->flags
))
3718 return ZONE_RECLAIM_NOSCAN
;
3720 ret
= __zone_reclaim(zone
, gfp_mask
, order
);
3721 clear_bit(ZONE_RECLAIM_LOCKED
, &zone
->flags
);
3724 count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED
);
3731 * page_evictable - test whether a page is evictable
3732 * @page: the page to test
3734 * Test whether page is evictable--i.e., should be placed on active/inactive
3735 * lists vs unevictable list.
3737 * Reasons page might not be evictable:
3738 * (1) page's mapping marked unevictable
3739 * (2) page is part of an mlocked VMA
3742 int page_evictable(struct page
*page
)
3744 return !mapping_unevictable(page_mapping(page
)) && !PageMlocked(page
);
3749 * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list
3750 * @pages: array of pages to check
3751 * @nr_pages: number of pages to check
3753 * Checks pages for evictability and moves them to the appropriate lru list.
3755 * This function is only used for SysV IPC SHM_UNLOCK.
3757 void check_move_unevictable_pages(struct page
**pages
, int nr_pages
)
3759 struct lruvec
*lruvec
;
3760 struct zone
*zone
= NULL
;
3765 for (i
= 0; i
< nr_pages
; i
++) {
3766 struct page
*page
= pages
[i
];
3767 struct zone
*pagezone
;
3770 pagezone
= page_zone(page
);
3771 if (pagezone
!= zone
) {
3773 spin_unlock_irq(&zone
->lru_lock
);
3775 spin_lock_irq(&zone
->lru_lock
);
3777 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
3779 if (!PageLRU(page
) || !PageUnevictable(page
))
3782 if (page_evictable(page
)) {
3783 enum lru_list lru
= page_lru_base_type(page
);
3785 VM_BUG_ON_PAGE(PageActive(page
), page
);
3786 ClearPageUnevictable(page
);
3787 del_page_from_lru_list(page
, lruvec
, LRU_UNEVICTABLE
);
3788 add_page_to_lru_list(page
, lruvec
, lru
);
3794 __count_vm_events(UNEVICTABLE_PGRESCUED
, pgrescued
);
3795 __count_vm_events(UNEVICTABLE_PGSCANNED
, pgscanned
);
3796 spin_unlock_irq(&zone
->lru_lock
);
3799 #endif /* CONFIG_SHMEM */